The two-process distinction in apparent motion

Copyright 1989 by the American Psychological Association, Inc. 0033-2909/89/S00.75 Psychological Bulletin 1989, Vol. 106, No. 1, 107-127 The Two-Process Distinction in Apparent Motion This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. J. Timothy Petersik Ripon College Traces the historical development of the notion that two processes mediate apparent motion percepts, examines evidence regarding their existence, and summarizes associated characteristics. The short-range process is assumed to reflect the activity of low-level directionally selective motion detectors, have a relatively small spatial integration range, and be favored by short stimulus durations and Intel-stimulus intervals. The long-range process is thought to reflect higher order perceptual activity, match stimulus elements over relatively large retinal distances, and be favored by longer stimulus durations and Intel-stimulus intervals. Criteria for associating different percepts with functionally different processes are advanced and applied. The theoretical status of the two-process distinction is examined, and a heuristic model of motion perception is presented. Apparent motion (AM) is the experience of motion that occurs when at least two spatially separated stimuli are alternately clarify the context of their theoretical descriptions; to examine presented to an observer over time. Because it is obviously an illusion and because it raises a number of penetrating questions temporal properties; to evaluate theoretical orientations toward explaining them; and to address the question of when two re- about psychological processes (e.g., how can the direction of motion be correctly imparted to a stimulus occurring later in functionally distinct processes. time from a stimulus occurring earlier in time without any prior information about the forthcoming location of the second stim- Because studies of AM frequently involve the simultaneous manipulation of variables in the distal spatial, proximal retinal, ulus?), the phenomenon of AM has fascinated researchers for well over 100 years. Until relatively recently, however, the as- temporal, and feature domains, it is often difficult to assign re- the evidence regarding their existence; to describe their spatio- lated but different percepts can be regarded as arising from sumption has been tacitly made that all instances of AM arise search reports to neat organizational categories. Therefore the categorization of research articles under the section headings in from the same fundamental psychological process or set of pro- this article has been somewhat arbitrary. Wherever possible, I cesses. In the last 15 years, growing recognition of the existence of at least two distinct processes that mediate the perception of AM has led to somewhat of a shift of orientation in the study have referred the reader to other sections of this article that bear on research reviewed in a given section. of visual motion perception. Prior to the mid-1970s, a main Historical Development of Concept of Two AM Processes focus of research in AM was to determine whether real movement (RM) and AM represented two aspects of the same psychological and physiological phenomenon or whether they con- The occurrence of distinctly different qualitative experiences of movement that correspond to what are now called short- and stituted different, independent processes (e.g., Clatworthy & Frisby, 1973; Kaufman, Cyrulnick, Kaplowitz, Melnick, & Stof, 1971). Since the putative division of AM into short-range long-range movements was discovered early in the study of AM. For example, in one of his experiments, Ternus (1938) had sub- and long-range AM, however, the goal of much research has jects view successive presentations of the two stimulus frames been to elucidate the properties of and relationships between shown in Figure 1. As a member of the Gestalt movement, Ter- these two processes, as well as to determine which, if either, process mediates the perception of RM. Whereas once research titles referred only to apparent motion, more and more they nus was interested in knowing whether the perceived AM would be of a single group of elements moving back and forth, in which case the phenomenal identity of the stimulus elements would be based on their relative positions within the group, or include qualifiers regarding the type of AM studied (e.g., see the titles of Prazdny, 1986a, 1986b). Given the strong influence of the putative discovery of two AM processes on research in the whether the perceived movement would only involve the outer elements while the central elements remained stationary, in which case the phenomenal identity of the stimulus elements area of perception, this review was undertaken with the following objectives in mind: to present the historical background underlying the discovery of two processes in AM and thereby to would be based on spatial location. In raising the problem of the determination of the phenomenal identity of stimulus elements in AM, Temus was the first to study what is known today as the correspondence problem (e.g., Oilman, 1979), or the ques- I am grateful to La Verne A. Toussaint and Bonnie Wolff for their careful preparation of this manuscript and to two anonymous reviewers for their suggestions regarding improvement of an earlier draft of this article. Correspondence concerning this article should be addressed to J. Timothy Petersik, Department of Psychology, Ripon College, P.O. Box 248, Ripon, Wisconsin 54971. tion of how individual elements in multielement AM displays become perceptually matched with their corresponding partners across frames. Although the visual system's solution to the correspondence problem was not completely elucidated by Ternus (see the following paragraphs) and has by no means been adequately characterized yet (although much progress has been 107 108 J. TIMOTHY PETERSIK This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. Time Time Figure I. Schematic example of the Ternus (1938) stimuli showing the two frames,/ andf2, as they are presented over time (top) and showing the sequence of presentation of the frames and interstimulus intervals (ISls) over time (bottom). made by Ullman, 1979, 1980), recent thinking suggests that alternative strategies may be used under different conditions. These strategies are represented theoretically by what are called the short- and long-range processes in AM. Regarding the stimuli of Figure 1, Ternus (1938) discovered that the overriding percept was of a group of elements in AM, a finding that was congenial to his orientation as a Gestalt psy- in the two frames in cooperation with a global process that performs a correlation on the entire arrays. Because AM also involves a perceptual matching of corresponding elements in two frames (separated in space and time instead of separated by a retinal disparity as in stereopsis), Julesz (1971) and others (e.g., Anstis, 1970; Braddick, 1973) attempted to produce a perceptual segregation of the correlated chologist. However, he also noted that end-to-end movement of the outer stimulus elements could be obtained when subjects directed their attention to the central elements. After Ternus' regions in random-dot frames by alternating them tachistoscopically. The result was that within certain ranges of alternation and displacement, the displaced region of a random-dot cine- discussion of phenomenal identity in terms of Gestalt organiza- matogram was also perceptually segregated. The similarities tional principles, there was little further investigation of the phenomenon, and for a long while, there was virtually no devel- and in cinematograms were obvious, and Julesz again applied between the segregations observed in random-dot stereograms opment of formal explanations of the (at least two) subjective the notion of local and global processes to explain the phenome- differences experienced with similar AM displays. non. By this account, in AM of random-dot stimuli a local process may perform matches of distinctive features (clusters of The first attempt to formally identify different organizational processes in multielement AM arose after Julesz (1971) described local and global processes in the perception of random- elements or blobs) prior to a global correlation process. The global correlation process, aided perhaps by the local process, dot stereograms. The typical stimulus studied by Julesz consisted of two frames (« X n matrices), the elements of which performs a point-by-point comparison of elements in a number of displacements in order to arrive at a best fit, thereby solving were made either black or white by random determination. The two frames were identical except for a small, square region that the correspondence problem and permitting the perceptual seg- was displaced in one frame in relation to the other, establishing a binocular disparity between corresponding regions. Although no objective contour was present in either frame individually, when combined stereoscopically, the two frames gave rise to the percept of a square region appearing in a depth plane different from the background elements. The question posed by Julesz was how the perceptual system correctly matched corresponding elements and avoided the numerous possible false matches (this is the stereoscopic version of the correspondence problem) regation of the displaced region. Braddick (1974) conducted an investigation of the spatial and temporal conditions under which a correlated and displaced region of alternating random-dot displays with uncorrelated backgrounds can be segregated, and he arrived at a number of conclusions: (a) The maximum displacement (rf max ) between frames that still yielded perceptual segregation was an absolute value (15 arc min) and not dependent on the size of the stimulus elements; (b) perceptual segregation decreased as the interstim- in the area of the displaced square. His answer was that the perceptual segregation of corresponding dots in the two frames ulus interval (ISI) between frames increased from 10 to 80 ms; (c) segregation could not occur when the stimulus frames were presented dichoptically; and (d) a light-filled ISI abolished per- could occur by a local process that matches distinctive features ceptual segregation (Braddick, 1973). Because classical AM be- 109 TWO-PROCESS DISTINCTION IN APPARENT MOTION tween well-defined stimulus contours can occur with large dis- was seen to rely on the activity of a short-range process, whereas 1SI, and diehoptically, Braddick (1974) suggested that the perceptual segregation obtained with alternating random-dot stim- group movement relied on a long-range process.' uli was due to the operation of a low-level motion-detecting process with a limited spatial range, perhaps based on directionally selective neurons in the visual system. Braddick contrasted this short-range process with a higher-order long-range process that presumably operates in classical phi phenomena; however, the long-range process was not thought to be able to segregate correlated regions in random-dot displays. There are some similarities between Braddick's (1974) short- This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. for the classical phi phenomenon. Hence, end-to-end movement placements, over a wide range of ISIs, with a light- or dark-filled Discriminating Functionally Different Perceptual Processes Braddick's (1974) argument for the existence of a short-range process in AM was based on his finding that the segregation of correlated areas in alternating random-dot frames occurred over limited spatial and temporal ranges and limited conditions of stimulus presentation. Such conditions were narrower than and long-range processes and Julesz's (1971) global and local processes, respectively. For example, both the short-range pro- those under which the phi phenomenon can ordinarily be ob- cess and the global process seem to operate on the basis of corre- rate processes that mediate different percepts of AM. In the ab- lation. Also, both the long-range process and the local process sence of independent evidence for two processes in AM, it could he argued that segregation of correlated areas in alternating operate by the identification of easily recognizable stimulus served. In and of itself, such a finding cannot offer proof of sepa- forms or contours. However, Braddick's suggestion that the random-dot displays represents a limiting condition or special short-range process is low level (presumably operating earlier than the higher-order long-range process) is incompatible with case of ordinary phi movement. The effort to discriminate the Julesz' notion that the local process "does some preprocessing before correlation" (Chang & Julesz, 1983a, p. 639). This raises the question of whether the short- and long-range processes operate serially or in parallel (see the section Failures to Find Two Processes). Shortly after Braddick's (1973, 1974) studies of the conditions under which the correspondence problem is solved for alternating random-dot displays, Pantle and Picciano (1976) and Petersik and Pantle (1979) conducted investigations of the conditions under which the correspondence process leads to endto-end and group movement percepts with the Ternus (1938) stimuli shown in Figure 1. Pantle and colleagues showed that under certain conditions of presentation, the Ternus display is bistable, that is, with prolonged viewing perception of the display alternated between end-to-end and group movement, reflecting competition between the short- and long-range processes when they were equally, or nearly equally, stimulated. Changes to stimulus conditions other than those favoring bistability altered the relative frequency of the two percepts: Pantle and Picciano found that end-to-end movement was obtained less frequently as ISI increased up to a limit of about 80 ms, that dichoptic viewing eliminated the end-to-end percept altogether (i.e., group movement was always perceived), and that the contrast reversal of the stimuli in one frame also eliminated endto-end movement. Additionally, Petersik and Pantle found that both the end-to-end and group movement percepts could be selectively adapted and that a light-filled ISI eliminates end-toend movement. Pantle and Petersik (1980) also showed that displacements of the overlapping elements in the Ternus display of 16 arc min were sufficient to eliminate the perception of end-toend movement. Because of the similarities to Braddick's (1973, 1974) findings, Petersik and Pantle (1979) and Pantle and Petersik (1980) concluded that the end-to-end movement percept (specifically, the percept of stationarity among the overlapping elements) occurred under the same conditions that favored segregation of correlated and displaced regions in alternating random-dot displays. Similarly, any conditions that prevented segregation in random-dot displays also prevented end-to-end movement in the Ternus display. Petersik and Pantle also identified group movement with the same AM process responsible bases for the short- and long-range movement percepts raises the more general question of when it is valid to postulate that two different percepts in the same phenomenological domain (e.g., motion, color, size) arise from functionally different processes. The question is theoretically important because it is often the case that stimulus differences simply lead to different outputs, and hence percepts, from a single integrated process. For example, few researchers today would argue that the different experiences associated with the colors yellow and green arise from functionally different processes; rather, they are produced by the relative activity produced in the three cone systems of the retina by stimuli differing in their spectral compositions. Criteria Although this list may not be exhaustive, evidence converging on a two-process perceptual interpretation should consist of the following: (a) The two perceptual experiences should be at least locally mutually exclusive (i.e., the same elements of a display should not simultaneously partake in both percepts); (b) the temporal changeover from one percept to the other should be abrupt, not gradual; (c) a changeover in perceptual experience should be observed in a variety of different displays; (d) such changeovers should occur under similar stimulus conditions; (e) the phenomena associated with each process should obey different principles, that is, no single psychophysical relationship should be able to describe both perceptual phenomena; (f) if the mutually exclusive percepts are thought to be in competition, they ought to be selectively adaptable; and (g) the perceptual experiences obtained with different displays should conform to the predictions of a single theoretical framework that describes the relationships between them. 1 Petersik and Pantle (1979) initially identified end-to-end movement with a hypothetical process they called t and group movement with a process they called y. Because their descriptions of the properties of those hypothetical processes are the same as the characteristics commonly associated with the short- and long-range processes, respectively, and in an effort to reduce the extra theoretical terminology, their terms are not used further in this article. 110 J. TIMOTHY PETERSIK Evidencefor Existence of Short- and Long-Range Processes point and its partner, other sequences of dots are being plotted and erased elsewhere on the screen, producing continuous dynamic events. The task of Morgan and Ward's subjects was to Displays designed to discriminate short- and long-range motion satisfy the first two of the preceding criteria: With randomdot displays, either observers experience coherence of the displaced patch or they do not; with the Ternus (1938) display, they see either group movement or end-to-end movement. Furthermore, under conditions that have been studied to date, perceptual changeovers always occur abruptly and effortlessly. A number of experimental reports have also satisfied the second three of the preceding criteria: Work with the Ternus display by Pan- This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. tie and Picciano (1976), Petersik and Pantle (1979), Pantle and Petersik (1980), and others has demonstrated a changeover in perceptual experience (from end-to-end movement to group movement, and vice versa) that occurs under conditions similar to those under which segregation of correlated random-dot indicate the direction of the overall flow of motion, which requires that subjects correctly match stimulus dots with their subsequent partners. The findings of the study were that motion flow in a single (correct) direction could only be perceived when the spatial separation between dots was no greater than 18 arc min and their temporal separation no greater than about 60ms in normal room light (similar boundaries were found with a somewhat different display by van Doom, Koenderink, & van de Grind, 1985). Otherwise, incoherent motion is perceived. Again, the limiting conditions for motion flow mimic those for Braddick's (1974) short-range process, and beyond those conditions the percept changes dramatically. Hildreth (1983) described a number of new displays that produce different perceptual results depending on whether the con- clusters occurs and is lost. Gerbino (1981, 1984) described yet ditions of stimulation favor the short- or long-range process. In another display that shows short- and long-range phenomena. In Gerbino's two-frame display a pair of neighboring, oppositely one demonstration, for example, a set of thin, equally spaced (16 arc min apart) black bars were presented on a grey back- pointing isosceles triangles (i.e., triangles whose bases face one another) are presented briefly. After a short ISI, only one of the original two triangles reappears. Gerbino found that with short ground for 33 ms. After an ISI of only a few milliseconds, the bars were shifted 4 arc min to the right, and the central segment (about 30%) of each bar reversed its contrast (i.e., the central ISIs (e.g., 5-30 ms) the dominant percept was of the disappearance of one of the triangles, whereas with longer ISIs (e.g., 40- conditions, the outer (black) portions of the bars appeared to 80 ms) the perceptual experience was of a three-dimensional segment changed from black to white). Under these short-range rotation of a single triangle from side to side. Furthermore, move to the right, whereas the central segments appeared to move to the left. However, when the stimulus duration was dichoptic viewing of the triangle that appeared in both frames of the display eliminated the percept of disappearance; threedimensional rotation was reported with all ISIs. Because of the similarities to the data of Pantle and Picciano (1976), Gerbino raised to 150 ms and the ISI to over 50 ms, conditions that favor the long-range process, only the outer segments were perceived as moving coherently; the central segments only flickered between black and white. (1984) concluded that the percept of three-dimensional rota- Recently, Dick, Ullman, and Sagi (1987) showed that the stimulus conditions that have favored short- and long-range AM tion was due to the activity of a high-level long-range process, whereas the percept of local disappearance of a single triangle was due to the filtering of a low-level short-range process. Bell and Lappin (1979), Bishof andGroner (1985), and Petersik (1989) presented observers with displays consisting of circu- also produce differences in the attentive abilities of subjects. The task of the subject was to detect the occurrence of motion involving a single dot pair embedded in an array of other stimulus elements that in one experiment were stationary and in an- lar areas randomly filled with dots. From the first frame to the other experiment simply appeared or disappeared. Dick et al. second, the dots were rotated by a variable amount around the center of the circular patch. Bell and Lappin (1979) and Petersik found that reaction time was shorter in response to short-range motion than to long-range motion and that reaction time to (1989) found that with relatively small rotations and short ISIs short-range motion did not increase as the number of back- (about 0-50 ms) the percept is of coherent rotational movement of the stimulus elements. Beyond the limiting rotation and ISI, ground elements increased. However, reaction time to long- incoherent motion of the elements is perceived. Bishof and Groner's results were similar. In addition, Petersik (1989) found range motion increased linearly as the number of background dots increased. Dick et al. interpreted their results as showing that the coherent rotational motion was difficult to obtain with that the short-range process operates in a preattentive, parallel manner; that is, motion is coded directly and independently of any rotation or ISI when the viewing of the two frames was events in the background. Long-range processing was thought dichoptic. These results once again suggest that the rotational to operate in an attentive, serial manner; that is, motion results from the conjunction of disappearance and appearance detec- AM is due to a local matching of elements performed by a relatively peripheral short-range process. The incoherent motion observed with large rotations and long ISIs is likely to be due to random matches of stimulus elements made over distances favorable to both short- and long-range processes. Morgan and Ward (1980) developed an oscilloscopic display that produced what they called motion flow in dynamic noises. In this display a set of randomly positioned dots is plotted one at a time. Each dot is subsequently erased and replotted after a specific interval a constant distance from the original dot. What makes this display unlike those used in classical AM studies is that during the interval between the presentation of any single tion and reflects the shifting of an "attention aperture" over parts of the visual field. In addition to the evidence supporting the existence of two processes in AM, recent work has shown that each process obeys different psychophysical relationships. One of these is the different reaction-time functions obtained by Dick et al. (1987) for short- and long-range movement. Larsen, Farrell, and Bundeson (1983) inferred the existence of short- and long-range processes in a simple AM study on the basis of different psychophysical functions. In their experiment, the stimulus onset asynchrony (SOA) required to obtain optimum AM between This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. TWO-PROCESS DISTINCTION IN APPARENT MOTION two alternating stimulus elements was measured as a function of the spatial separation between those elements. Larsen et al. found that although SOA consistently increased with spatial separation, two linear functions were required to account for the changes in the data. For separations up to about 15 arc min (which is close to Braddick's [1974] putative d^f for the shortrange process), SOA increased with a slope of 265 ms/degree, whereas for greater separations, SOA increased with a slope of 18 ms/degree. Furthermore, Larsen et al. showed that changing viewing distance raised the slope for the long-range function but not for the short-range function. They concluded that the shortrange process is more peripheral and responds to retinal separations between stimuli, whereas the long-range process is more central and responds to perceived separations. Braddick (1980) showed a similar split in the functions describing an AM sensation over different ranges of spatial separations. In this case an annulus divided into a number of black and white sectors is rotated clockwise and exposed in steps of one fifth of a period. If the ISI of the display is illuminated, the display becomes ambiguous, with the percept alternating between clockwise and anticlockwise movements. Braddick plotted the SOA required to produce reversed (anticlockwise) motion for 33% of a viewing period as a function of the spatial separation between successive exposures of annulus sectors. He found that SOA increased linearly on a log-log plot up to a separation of about 30 arc min. Beyond this point, the data rose rapidly. Braddick suggested that the 30-min breakoff point was the displacement limit of the short-range process; because viewing was somewhat peripheral, dmax was slightly larger than that obtained with random-dot stimuli (see the section Limited spatial range of short-range process?). The sixth criterion for discriminating functionally different short- and long-range processes—namely, that if the two processes are considered to be in competition, adaptation to one should increase the frequence of reports of the other—was satisfied by Petersik and Pantle (1979). Using the Ternus (1938) display, they showed that adaptation to short-range (end-to-end) movement increased reports of long-range (group) movement over a range of ISIs from 20 to 70 ms. Adaptation to long-range movement produced similar increases in reports of short-range movement. Experimental investigations have been shown to satisfy the first six of the initial seven criteria proposed as necessary for distinguishing functionally different perceptual processes. What remains is to show that the two putative processes fit into a single theory of AM. As of this writing, there is no single theory of AM that completely explains all the phenomena of both short- and long-range movement as well as their relationships. Lappin and Bell (1976) suggested a cross-correlation process as a model of the process producing motion coherence in alternating random-dot displays, and Baker and Braddick (1985b) implemented the model with Reichardt-like detectors. In her monograph concerning the computational approach to modeling the visual interpretation of motion, Hildreth (1983) described the need for computational and perceptual studies of both short- and long-range motion. However, her own theoretical work is limited to showing that in complex displays the computation of local velocity vectors (which can in turn be used to model the true global velocity field of a moving scene) can be achieved by the short-range process. Part of the reason for the 111 lack of theoretical integration is due to the numerous unsolved questions regarding each process. Additionally, there is ample competition among theorists to simply account for the psychophysical relationships obtained in a large body of literature without trying to relate them to two putative processes; this has resulted in little attempt to theorize about the function and evolutionary history of the two AM processes. Nonetheless, a number of theoretical suggestions aimed at accommodating the short-range/long-range distinction have been made in the literature. Some of these are reviewed in the next section, and attempts to evaluate them are made throughout this article. Theoretical Points of View Peripheral/Central Processes A number of researchers have suggested that the short-range process represents the activity of peripheral, early perceptual processing, whereas the long-range process reflects the activity of more central processing. The studies of Pantle in which gratings have been abruptly shifted to produce AM (Pantle, Eggleston, & Turano, 1985; Pantle & McCarthy, 1988; Pantle & Turano, 1986) have been directed at demonstrating this distinction. Pantle postulates a hierarchical model in which luminance matches are made to establish correspondences and generate motion signals at a low-level stage of processing and in which "pattern primitive" tokens are matched at a higher-level stage, also generating motion signals. Such a model is consistent with the results obtained with random-dot displays and the Ternus (1938) display, especially those suggesting that the short-range process cannot operate with dichoptic viewing and is sensitive to the sign of the contrast of the displays (Pantle & Picciano, 1976; Petersik & Pantle, 1979). The major obstacle facing this point of view is in explaining how and why the output of the lower level stage bypasses that of the higher level stage when conditions favor short-range movement. Motion Detectors and Perceptual Processes Consistent with the peripheral/central distinction is the point of view, only partially developed, that short-range motion may reflect the output of neural motion detectors, whereas longrange motion may be mediated by subtler correspondence rules (Anstis, 1980; Braddick &Adlard, 1978). Accordingto this perspective, because the short-range process has a limited spatial range and is typically not observed with dichoptic AM, it ought to show properties similar to low-level retinal motion detectors (like those discovered by Barlow & Levick, 1965). On the other hand, the long-range process seems to group elements in AM in ways that are ecologically meaningful. In so doing, it tolerates more stimulus variability than does the short-range process. For example, the long-range process ignores stimulus differences in color and shape (Kolers & von Grunau, 1976), contrast (Pantle & Picciano, 1976), and orientation (Pantle & Petersik, 1980). Furthermore, in resolving ambiguous input, the long-range process may alternatively generate motion percepts in two-dimensional or three-dimensional space (Kolers & Green, 1984). Finally, characteristics of long-range AM show a great deal of individual variability and are susceptible to practice effects (Finlay, Manning. & Fenelon, 1987). However, despite the continuing 112 J. TIMOTHY PETERSIK research trend aimed at establishing correspondence rules of the long-range process (some examples of which are reviewed in the section Correspondence rules for long-range AM), there has been little success in establishing a theoretical framework in which to interpret the variety of phenomena reported. This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. Relationship to Pattern Persistence Consistent with the peripheral/central distinction is the theory currently being developed by Breitmeyer (Breitmeyer & Ritter, 1986a, 1986b; Ritter & Breitmeyer, 1987) that the shortand long-range motion percepts obtained with the Ternus (1938) display can be explained, in part, by pattern persistence. The argument is basically that conditions that enhance pattern persistence will also enhance the perception of stationarity of the overlapping elements in the Ternus display. On the other hand, conditions that reduce pattern persistence will make it difficult to perceive the stationarity of the overlapping elements and a process of temporal differentiation will lead to segregation of the groups of elements in the two frames and ultimately to the perception of group movement. Thus, for example, because there is an inverse relationship between stimulus duration and pattern persistence, it is expected that longer stimulus durations would reduce end-to-end movement and increase group movement. This is what was found by Petersik and Pantle (1979). The visual persistence hypothesis also accounts for Pantie and Picciano's (1976) finding that illuminating the ISI of the Ternus display prevents end-to-end movement; visual persistence is known to decrease as light adaptation increases (Breitmeyer & Ritter, 1986a). Breitmeyer and Ritter (1986a) also showed that smaller stimulus elements favored end-to-end movement in comparison with larger elements. This was explained as being due to the fact that pattern persistence increases with increases in spatial frequency. They also showed that reports of end-to-end movement varied inversely with retinal eccentricity up to at least 2 degrees, which is consistent with the finding that pattern persistence decreases with eccentricity (Breitmeyer & Ritter, I986b). Finally, Ritter and Breitmeyer (1987) found, in contradiction to Pantle and Picciano (1976) and Braddick and Adlard (1978), that reports of end-to-end movement can be obtained with dichoptic viewing of the Ternus (1938) stimuli if the conditions of presentation otherwise optimize pattern persistence. As Breitmeyer and Ritter (1986a, 1986b) noted, the pattern persistence hypothesis does not account for all the data previously obtained with the Ternus (1938) display. Whereas Petersik and Pantle (1979) found that decreasing the contrast of the stimulus elements in the Ternus display favors group movement, the pattern persistence hypothesis predicts that lower contrast should favor end-to-end movement. However, Petersik and Pantle varied contrast by varying the intensity of the stimulus elements on a background of constant luminance, not by simultaneously varying the intensity of elements and background to maintain a constant space-average luminance. Breitmeyer and Ritter (1986a) suggested that this may have increased the relative light adaptation at the lower contrast, thus favoring group movement. Also, Petersik and Grassmuck (1981) reported that increasing fundamental spatial frequency favors group movement, whereas pattern persistence increases with higher spatial frequencies, which should in turn favor end-to- end movement. However, as the spatial frequency of Petersik and Grassmuck's stimuli increased, the center-to-center separation of the elements decreased, and the overall area subtended by the display also decreased. It is possible that these factors may have overcompensated for the spatial-frequency effect predicted by the pattern persistence hypothesis and that under conditions that control them, increasing spatial frequencies may be associated with more end-to-end movement. Sustained and Transient Channels Petersik and Pantle (1979) initially suggested that the processes mediating group and end-to-end movement in the Ternus (1938) display parallel properties of transient and sustained channels, respectively. After this, Breitmeyer and Ritter (1986a) noted that because pattern persistence is associated with activity in sustained channels (Breitmeyer, 1984), sustained channels may also be associated with end-to-end movement. On the other hand, because pattern persistence is inhibited by activity in transient channels, group movement may be the outcome of activity in transient channels. Therefore, this perspective is consistent with the pattern persistence hypothesis and may provide a physiological basis for understanding both persistence phenomena and aspects of AM. Summary of Theoretical Perspectives Although the terminology may differ, the theoretical perspectives that have been presented in this section share a number of features regarding the two putative processes in AM. For example, most can be interpreted as suggesting that the short-range process occurs at an earlier, or more primitive, level of perceptual analysis than the long-range process. The evidence for this seems to be that short-range percepts are more easily disturbed than long-range percepts (e.g., by contrast, orientation, spatial separation, viewing, or temporal changes), suggesting a certain dependence on strict stimulus parameters. On the other hand, most of these perspectives imply that long-range processing is more flexible and perhaps more susceptible to cognitive influences. In this regard, it may be possible to rule out certain aspects of these approaches and, in so doing, arrive at a de facto theoretical synthesis. In the following sections, empirically derived characteristics of the short- and long-range processes are reviewed with the goal of focusing on salient theoretical conclusions. After that, a functional interpretation of the two-process distinction is offered, with an emphasis on its relation to understanding the perception of real movement. Characteristics of Short- and Long-Range AM Processes Table 1 shows a selective list of characteristics and limiting conditions that have been proposed for the short- and longrange processes. Many of these are drawn from summary papers by Anstis (1980) and Braddick (1980). Although salient citations are supplied, in most cases, supporting research comes from several sources. The list is necessarily incomplete, especially in terms of characteristics of the long-range process (for a review of characteristics of the long-range process, see Kolers, 1972). It should also be mentioned that the division of topics into characteristics of the short- and long-range processes is TWO-PROCESS DISTINCTION IN APPARENT MOTION 113 Table 1 Putative Characteristics of the Short- and Long-Range Processes in Apparent Motion Short range Study Anstis, 1980; Braddick, 1980; Pantle & Petersik, 1980 Braddick, 1980; Petersik & Pantle, 1979 Baker & Braddick, 1985b; Petersik & Pantle, 1979 Petersik & Grassmuck, 1981 Braddick, 1980; Pantle &Picciano, 1976 Braddick, 1980; Pantle &Picciano, 1976 This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. Braddick, 1980; Petersik & Pantle, 1979 Anstis, 1980; Braddick, 1980 Petersik & Pan tie, 1979 Anstis, 1980 Anstis, 1970, 1980; Lappin & Bell, 1976 Anstis, 1980 Dick, Ullman, & Sagi, 1987 Larsen, Farrell, & Bundesen, 1983 Cavanagh, Boeglin, & Favreau, 1985; Chang & Julesz, I983a Chang &Julesz, 1984, 1985;Julesz, 1971 Petersik, Hicks, & Pantle, 1978 Breitmeyer & Ritter, 1986a, 1986b Small spatial integration range (2ff or less)" Prefers small ISIs (<40-60 ms) Prefers shorter stimulus durations Prefers lower fundamental spatial frequencies Cannot operate with dichoptic stimulation Cannot operate with contrast reversal of stimuli ISI must be dim or dark Isoluminant colors not adequate stimuli (stimuli must be defined by luminance contrast) Requires identity or similarity of form between corresponding elements Stimulates neural motion detectors and produces motion aftereffects Based on point-by-point cross-correlation type of process Motion precedes edge extraction Preattentive (parallel processing) Responds to retinal separations May have two levels of processing: motion detection and figural segregation Is a global, cooperative process Output can serve as input to the long-range process Is at least partly based on visual persistence Long range Large spatial integration range (up to several degrees) Prefers longer ISIs Prefers longer stimulus durations Prefers higher fundamental spatial frequencies Operates both monoptically and dichoptically Contrast of stimuli largely irrelevant Luminance of ISI largely irrelevant Chromatic contrast or luminance contrast are adequate Form similarity irrelevant May not stimulate motion detectors and produces little motion aftereffect Based on detection of subtler (higher order?) correspondences Edge extraction precedes motion Attentive (serial processing) Responds to perceived separations Unknown number of levels of processing Is based on feature preprocessing N/A Inhibited by visual persistence Note. ISI = interstimulus interval; N/A = not applicable. " It is clear that researchers no longer consider the spatial limit of the short-range process to be fixed at 15'-20'. At the same time the short-range limit is orders of magnitude lower than the limit of the long-range process (Julesz & Schumer, 1981). somewhat arbitrary. If nothing else, current research shows that the two processes are complementary. In describing the bound- periment that made dmm artificially appear to have a fixed retinal limit. aries of one process, one is nearly always simultaneously studying limiting conditions of the other. Lappin and Bell (1976) may have been the slight difference in One reason for the difference between Braddick (1974) and the dependent measures used. However, in a similar experi- Characteristics of Short-Range Process Limited Spatial Range of Short-Range Process? Lappin and Bell (1976) were the first to criticize Braddick's report of a small displacement limit for segregation in alternat- ment, Chang and Julesz (1983a) found that although rfmail was indeed smaller for a motion coherence task (similar to that of Lappin and Bell), it still increased up to 25 arc min with increases in target size, suggesting once again that rfmax is not absolute. In addition, Chang and Julesz (1983a) determined that </„„ for detection of the direction of coherent motion increased ing random-dot displays and to provide contradictory data. In proportionally to the square root of the area of the displaced their own study of random-dot displays, Lappin and Bell found random-dot region (see also Nakayama & Silverman, 1984), that coherent motion detection was a declining function of dis- was not affected by peripheral viewing (up to 4 degrees), and placement but an increasing function of the number of ele- was invariant with changes in viewing distance (50 to 100 cm). ments contained in the display (specifically, motion discrimina- Interestingly, their "area effect" was only evident in the dimen- tion was roughly proportional to the square root of the number sion parallel to the direction of movement, suggesting that the of pattern elements). Furthermore, there was evidence that the displacement limit depended on the spacing of elements, with a smaller spacing producing more correct discriminations of short-range process is directionally selective. In other research, motion. Unlike Braddick (1974), who found that dm^ was a constant value, Lappin and Bell found a relative limit. They The £/„, was smallest to largest for high-frequency filtered, explained the discrepancy between their data and Braddick's as being due to the fact that as Braddick increased the displacement in his patterns, he also increased the number of display quency filtered, and low-frequency filtered patterns, respec- Chang and Julesz (1983b) showed that rfmai varied as a function of the spatial-frequency content of the random-dot patterns. medium-frequency filtered, unfiltered, low- plus high-fre- tively (cf. Nakayama & Silverman, 1984; Ramachandran & Anstis, 1983). The smallest values of rfma, were in the neighbor- elements while holding the display dimensions constant. In so hood of 5 arc min, whereas the largest were about 25 arc min. doing, they argued, Braddick introduced a confound to his ex- To account for their data, Chang and Julesz (1983b) suggested This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. 114 J. TIMOTHY PETERSIK. the following psychophysical relationship: o?mlul/grain size = k, a constant. Although it is clear that the small dmt!t found with the high spatial frequencies is a function of the short-range process, it remains to be determined whether the larger dmm found with the low spatial frequencies is a function of the long-range process. Therefore, it is not clear whether these results are consistent with Braddick's (1974) postulated absolute retinal limit for the short-range process. However, Chang and Julesz (1983b) replicated their earlier finding that rfmax increased with target area and also found that dm^ was invarient with retinal eccentricity up to 4 degrees. Thus, they favored the interpretation that there was no absolute limit to the short-range process. Using different stimuli and a somewhat different dependent measure, Burr, Ross, and Morrone (1986) arrived at similar conclusions. An increase in </„,„ with increases in target area is one of the more enduring effects found in the study of the short-range process. Figure 2 shows the results from four different studies of the short-range displacement limit in random-dot stimuli. Despite differences in criteria and parameters of stimulus presentation, the data all come close to being described by the same ogival function. Some authors (e.g., Chang & Julesz, 1983a; Petersik, Pufahl, & Krasnoff, 1983) have interpreted the area effect as evidence against a fixed retinal limit of the short-range process. In view of the criticisms just cited, particularly the area effect, Baker and Braddick (1982a) undertook to explain the effects of target area and element density on dmm. To address the earlier criticism of Lappin and Bell (1976) regarding the change in the number of patch elements as spacing varied, Baker and Braddick varied element density so as to maintain an equal number of elements in the displaced patch for all element spacings. In summary, they found that (a) dot density had no effect on dm* (thereby apparently invalidating the criticism of Lappin & Bell, 1976); (b) dmm increased as patch area increased; and (c) dmx became larger as the target patch was presented more peripherally (a finding replicated by Baker & Braddick, 1985a, in which eccentricities up to 10 degrees were studied. This is well beyond the 4-degree range used by Chang & Julesz, 1983a.). Baker and Braddick (1982a) concluded that the increase in dm* with target area is due to the invasion of greater retinal eccentricities. Their failure to find an effect of element density was interpreted as evidence that </ma)1 is not determined by the number of elements displaced from frame to frame. For Baker and Braddick, </mas was better interpreted as a function of retinal angle than as the absolute number of element spaces used in displacement. Furthermore, because a cross-correlation model (e.g., Lappin & Bell, 1976) predicts a rfmax expressed in terms of the number of elements displaced, Baker and Braddick believed that their results reflected the operation of small direction-selective receptive fields more than a cross-correlation process. This interpretation was further supported by Baker and Braddick (1982b). Using a variation of the Ternus (1938) display, Petersik et al. (1983) argued that there was no fixed retinal limit to the shortrange process. In a modified replication of an experiment by Pantle and Petersik (1980), Petersik et al. replaced the dots in each frame of Figure 1 with long vertical bars of variable width. They also displaced the bars of one frame in relation to the other by 25-100% of the bar width. They subsequently showed that a single psychophysical function related the temporal equilibrium point between end-to-end and group movement to displacement only when displacement was expressed in relation to bar width, not when it was expressed in terms of retinal angle.2 Thus, with large bars (.72 degree), retinal shifts of .72 degree elicited good short-range movement, provided that ISI was short enough (about 12 ms). Petersik et al. also showed that in addition to spatial displacement, dmm depends on the temporal parameters of presentation. Subsequently, Baker and Braddick (1985b) replicated their studies of </max in random-dot displays, varying stimulus duration and ISI as well as displacement. The results were complex, but it was clear that dmm did change with both stimulus duration and ISI. From their data, Baker and Braddick concluded that part of each of two random-dot exposures must occur within the same roughly 40-ms interval in order for the shortrange process to permit the identification of motion direction. The balance of the evidence reviewed thus far suggests that dm** can indeed be described in terms of a retinal limit that changes as a function of the part of the retina stimulated. However, the data of Figure 2 and those of Petersik et al. (1983) show that rfraax must grow at a relatively constant rate across the retina. Petersik et al. (1983) questioned the theoretical usefulness of an absolute limit that changes smoothly with target area and retinal eccentricity. Although the short-range process shows aspects of directional selectivity, the variability of dmm over conditions suggests that it cannot be readily identified with retinal motion detectors. Recent research by Ritter and Breitmeyer (1987) also questions the relation between retinal motion detectors and short-range AM. In review, Table 2 shows estimates of (/ma, derived from a number of studies using a variety of stimuli, conditions of presentation, and response criteria. As can be seen, rfmax depends to a great extent on the stimulus conditions under which it is studied. For this reason, and as noted previously, it is questionable whether it is useful to consider dm!a as reflecting a fixed integration limit. The data of Table 2 also raise the question of whether all estimates of dmm reflect the activity of a single shortrange process (see the next section). Response of Short-Range Process to Color Contrast: Evidence for Two Levels? Ramachandran and Gregory (1978) showed that segregation of correlated patches cannot be obtained in equiluminous random-dot displays. In their displays, random dots of black and white (which provide luminance contrast) were replaced by red and green dots of equal luminance (which provided color contrast only). Their findings suggested that the short-range process is "color blind" and cannot make use of color contrast for the purpose of segregating correlated patches of random dots over 2 Temporal equilibrium point refers to the empirically determined ISI at which group and end-to-end movement are reported on 50% of all trials, that is, that ISI at which the competing sensations are equally strong. Thus, a large temporal equilibrium point reflects the fact that the short-range process (expressed as end-to-end movement) dominated over much of the range of ISIs tested. Similarly, a small equilibrium point means that the long-range process dominated over much of the range of ISIs. TWO-PROCESS DISTINCTION IN APPARENT MOTION 115 40 O flj 30 c I 20 Baker & Braddick (1982a) This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. Chang & Julesz (1983a) Q Chang & Julesz (1983b) 10 Nakayama & Silverman (1984) .1 1 10 100 Square Root of Patch Area (deg arc) Figure 2. Data from four experiments showing the variation in dm% as a function of the size of the displaced area in random-dot stimuli. (Data shown here were chosen for the ease with which points could be estimated from originally published reports and for the ease with which the original data could be converted to the format used in this graph. In cases for which the original reports showed data from more than one subject individually, the subject whose data were plotted here was chosen randomly.) time. Cavanagh, Boeglin, and Favreau (1985) optimized temporal conditions favoring the short-range process and found that a percept of coherent motion could be obtained under equilu- was 100% correlation of displaced elements in two opposite directions. To these patterns, varying amounts of biasing dots (i.e., dots displaced in one direction only) were added. The minance, whereas perceptual segregation was lost. This suggested that some short-range motion detection occurs on the finding was that the number of biasing elements needed to be basis of color contrast only. Similarly, efforts to display the Ter- of coherent motion in the biased direction only. Slightly more biasing elements, 6-8%, were required to produce the percept nus (1938) stimuli under equiluminance in the Ripon College Department of Psychology laboratory have shown that end-toend movement reports are reduced but not entirely lost. These results are consistent with the physiological findings of Livingstone and Hubel (1987, 1988), which indicate that the majority of motion detection occurs in the color-blind "magno system"; only 4-6% to disambiguate the display and produce percepts of coherent motion in otherwise uncorrelated random-dot patterns. These results are best explained if one assumes that local cooperative elements, perhaps motion detectors, interact within the structure of a global process, perhaps a correlationlike computational process. On the basis of a third experiment, Chang however, the color-sensitive "parvo system" indeed shows some and Julesz (1984, 1985) were able to determine that the average response to movement. Thus, consistent with a suggestion made cooperative "neighborhood" on the retina was about 15 arc min.3 by Chang and Julesz (198 3a), there may be more than one kind of short-range process: one associated with the magno system and responsible for coherent motion and perceptual segregation Williams and Sekuler (1984) also reported a cooperative percept of coherent global motion produced by local motion analy- and one associated with the parvo system and responsible for sis. In their random-dot displays (with dot densities in the range coherent motion only. of .2-1.6 dots/degree2), each dot was made to move in a random direction, within constraints, from frame to frame. The step Global Cooperativity of Short-Range Process To account for the binocular segregation of correlated and displaced regions in random-dot stereograms, Julesz (1971) postulated a cooperative global process in stereopsis that was modeled by the nonlinear coupling of neighboring local processes (dipoles). In order to demonstrate that the coherent motion produced by the short-range process was also cooperative, Chang and Julesz (1984) presented observers with ambiguous random-dot displays; that is, from one frame to the next there taken by each dot in each frame was independent of the direction of its previous steps and of the directions taken by other dots in the display; however, the distance traveled by each dot on each step was constant. The direction taken by each dot on J It is coincidental that the figure 15' is found for the average cooperative neighborhood by Chang and Julesz (1984) and for the approximate displacement limit of the short-range process by Braddick (1974). The similarity between the two should not be taken as evidence supporting the existence of either or both. 116 J. TIMOTHY PETERSIK Table 2 Estimates ofd^ Obtained in a Variety of Studies This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. Study Type of display Comments Baker &Braddlck(1982a) Baker &Braddick(1985b) RD RD Iff to 1° Baker&Braddick(1985a) Braddick(1974) RD RD 4-lOff 15' Rotating sector disk Sine wave gratings in AM RD X)' 2<yto6" 4-51' Chang &Julesz(1983b) Spatial-frequency filtered RD 5-27' Chang &Julesz( 1985) Spatial-frequency filtered RD 12' to 1' Morgan & Ward (1980) Motion flow in dynamic noise RD with variable pause at one-half the jump distance Ternus(1938)display Square-wave bars in Ternus (1938) arrangement Sparse RD 18' Braddick(1980) Burr, Ross, & Morrone(1986) Chang &Julesz(1983a) Nakayama & Silverman (1984) Pantle& Petersik (1980) Petersik, Pufahl, & Krasnoff (1983) Ramachandran & Anstis (1983) van Doom, Koenderink, & vandeGrind(1985) 12^10' <16' 1 23% of bar width At least 1.1° Spatiotemporal correlation in RD noise 15' to 2' Increasing function of area up to 10° Inverted U-shaped function of ISI (5-100 ms); increasing function of stimulus duration when ISIs short; separately dependent on ESI and stimulus duration Increasing function of eccentricity up to 10° Independent of element size, field size, and size of displaced region Viewing somewhat peripheral Decreasing function of spatial frequency Increase with square root of area; invarient eccentricities to 4° Increasing function of area for all frequencies; decreasing function of spatial frequency up to 7.4 cycles/degree; independent of eccentricity up to 4° Decreasing function of spatial frequency content, asymptoting at about 1 5' — Increasing function of field size up to 20° Based on position perturbations of overlapping lines — May not have stimulated short-range process Lower value at fovea, higher value at 25° peripheral Note. RD = random-dot display; ISI = interstimulus interval; dashes indicate no data. each step was drawn from a uniform probability distribution of directions, and the range of directions was variable in the studies. Despite the freedom of movement of the dots in the ensemble, coherent unidirectional motion was obtained (under certain conditions) when the range of possible directions was less than 360°." Overall, the results suggested a "spatial pooling for responses of direction selective mechanisms that are tuned to the mean direction of the distribution" (Williams & Sekuler, 1984, p. 61). Although the evidence is only indirect, the results can be attributed to a globally cooperative short-range process.5 To be achieved, this would require a global cooperative process of the kind discussed by Julesz (1971). The results of the studies just mentioned are all in agreement with the notion that the short-range process can be both global and cooperative. However, it is likely that the short-range process sometimes operates in the absence of globalism or cooperativity. The random and incoherent motion that characterizes percepts when rfma, is exceeded in random-dot displays (or when two uncorrelated random-dot frames are alternated) is likely to A cooperative function of the short-range process has also been observed in the Ripon College Department of Psychology laboratory (Petersik, 1989). The display consisted of two frames of dots randomly positioned within a circular area. The dots in the first frame were rotated about the circle's center by some angular amount to produce the second frame. Given appropriate timing, a rocking percept can be obtained when the two frames are alternated, up to an angular displacement of 80° or more. To obtain a global rocking percept, dots throughout the area of the circle must be correctly matched (i.e., the correspondence problem must be solved) across frames. However, when a mask of sufficient size is placed over the center of the circular patch (where the distances between partner dots across frames are small regardless of the angle of rotation), the coherent rocking motion is lost; that is, dots near the perimeter of the circle cannot be correctly paired across frames. These results indicate that the solution of the correspondence problem near the center of the circular area determines the solution near the periphery. 4 Although unidirectional motion of the pattern could be obtained, observers also perceived the random local motions of the elements. This suggests a simultaneous separation of outputs from local and global motion analysis that may be similar to the simultaneous perception of local and global motions corresponding to correlated and displaced regions in alternating random-dot frames or similar to the simultaneous perception of stationarity of overlapping elements and long-range motion of the outer elements in the Ternus (1938) display. * Williams and Sekuler (1984) found that variations in dot density proved to have no effect on unidirectional motion for small step sizes (< 1 arc degree); this is consistent with the results of Baker and Braddick (1982a) for the short-range process. Williams and Sekuler also found that increasing the number of frames in a display aided the perceptibility of unidirectional flow, which is similar to data obtained by Nakayama and Silverman (1984) with the short-range process. Williams and Sekuler's finding of a relatively large upper limit to the cooperative (short-range?) process can be accounted for on the basis of the fact that the display stimulated broad regions of the retina, both central and peripheral, thus increasing u"ma,. TWO-PROCESS DISTINCTION IN APPARENT MOTION 117 be the product of matches made by the short-range process of its spatial range makes it difficult to assert that there is a fixed when no global correlation can be found. It is possible that retinal limit to the short-range process that reflects the localized when short-range motion displays global cooperativity, it is due receptive fields of retinal motion detectors. On the other hand, to a spatial pooling of motion detectors designed to produce perceptual segregation (e.g., the magno system?). A more prim- its responses to equiluminant stimuli and its frequently ob- itive short-range motion may be based on individual responses the short-range process occurs after some spatial pooling of di- of motion-detecting units (e.g., the parvo system?). rectionally selective motion detectors. This, in turn, could take served global cooperativity suggest that what is ordinarily called place in the magno system of the cortex (Livingstone & Hubel, This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. Spatiotemporal Influences on Short-Range Process 1987, 1988). At the same time, a more primitive short-range process may account for the incoherent motion observed in Braddick (1974) initially showed a slight dependence of the many random-dot displays. This process may rely on the output short-range process on ISI. That is, longer ISls (up to 80 ms) of directionally selective cells, may be based on luminance increased the number of errors on the random-dot segregation matches, and may have its cortical representation in the parvo task and decreased the clarity of the target's boundaries. Peter- system. sik and Pantle (1979) showed that the short-range motion percept obtained with the Ternus (1938) display (end-to-end movement) decreased monotonically with increases in both stimulus duration and ISI. Furthermore, there was no simple trade-off between the two parameters, so stimulus onset asynchrony was not the critical variable mediating either end-to-end or group movement. In all conditions of their experiments (with the ex- Characteristics of Long-Range Process Because of the relatively large interstimulus distances involved, the majority of articles that have claimed to have studied AM have actually focused on the long-range process alone. Thus, a comprehensive review of the characteristics of the long- ception of adaptation to group movement and very short stimu- range process is beyond the scope of this article (the interested lus durations), an ISI of 70 ms resulted in the perception of group movement (the long-range percept) on over 70% of all reader should consult reviews by Anstis, 1978, 1980, 1986). trials. Later experiments by Petersik and Grassmuck (1981) and Petersik et al. (1983) showed that the dependence of the short- However, inasmuch as the long-range process is now thought to complement the activity of the short-range process and because new details regarding its operation have recently been uncov- range (and long-range) process on ISI itself depended on the ered, this section presents a selective review of information that spatial-frequency composition of the stimuli. The three bars of helps to distinguish the characteristics of the long-range process the Ternus (1938) display were transformed into three cycles of a sine wave grating. Increases in the fundamental spatial fre- from those of the short-range process. quency resulted in gradual decreases in the critical ISI needed to produce equilibrium between short- and long-range motion Interaction With Real Movement (RM) percepts. This in turn was thought to reflect a weakening of the short-range process, because the long-range process was activated at progressively shorter ISIs. The suggestion was made that the short-range process prefers low spatial frequencies, whereas the long-range process prefers high fundamental spatial frequencies.6 Such a conclusion is at least consistent with the notion that rfmax increases as random-dot displays are filtered to progressively lower spatial frequencies (Chang & Julesz, 1983b, 1985; Nakayama & Silverman, 1984). The finding also suggests that the temporal integration period of neural units whose receptive fields are assumed to underlie the short-range process increases as receptive field width increases. In a study in which the stimulus durations of both the first and second of two random-dot frames were variable, Baker and Braddick (1985b) found that at least part of both exposures must fall in the same roughly 40-ms interval in order for the short-range process to produce coherent motion. Nakayama (1985) also found evidence for an optimal temporal interval of about 40 ms for the short-range process, only for relatively high velocities (>20 degrees/s), however. Both studies suggest that the early stage of motion analysis, presumed to be the shortrange process, is very fast. Anstis (1980) and Braddick and Adlard (1978) suggested that the short-range process operates on the basis of stimulation of neural motion detectors, whereas the more cognitive long-range process does not. The logic for this conclusion seems to have arisen from the lack of existing evidence that the long-range process can generate motion aftereffects along with the fact that the short-range process can easily be modeled by the known characteristics of motion detectors. Burretal. (1986)essentially applied this logic in their study that showed that for small displacements, short-range AM can be made indistinguishable from RM. However, a number of more recent studies have seemingly demonstrated that the long-range process also responds, at least modestly, to RM. For example, Green (1983) showed that the long-range AM of grating patches was enhanced when the gratings within them were drifting in the same direction. Drift in the opposite direction decreased the frequency of AM reports. In a related experiment, Barbur (1981) optically combined a dim spot in RM with a similar target in AM, but at subthreshold levels of illumination (. 1 log unit below detectability). The positions of the discrete stimuli were about 2.5 degrees apart, so the AM was probably long range. The subjects' task was to adjust the RM stimulus to threshold illumination. Barbur found that AM in the same direction as the RM stimulus re- Implications for Theory In terms of the theoretical perspectives advanced in the section Theoretical Points of View, the characteristics of the shortrange process suggest two things. First, the apparent flexibility 6 However, in a study that appears to mainly implicate the long-range process, Ramaehandran, Ginsburg, and Anstis (1983) found a preference for low spatial frequencies in AM. This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. 118 J. TIMOTHY PETERSIK duced the threshold for detection of the latter almost linearly. Opposite-direction AM had no effect on threshold detection of RM (nor did continuously present stationary spots). Von Grunau (1986) used an AM stimulus similar to Green's (1983) to demonstrate that long-range AM can produce a motion aftereffect. Observers adapted to grating patches of either .5 or 1.0 cycles/degree (stimulus duration = 48 ms) that were alternately presented at a 4.2-degree separation with an ISI that was optimal for AM. The cycle rate was such that AM was seen in one direction only. The test stimulus was a continuously viewed sine wave grating patch that abruptly reversed its spatial phase. Ordinarily, such a stimulus is ambiguous: AM can be seen to either the left or the right with equal probability. Von Grunau reasoned that if AM produced direction-specific adaptation, then reports of AM in the phase-reversing grating should become biased in the direction opposite the AM. He found that AM indeed produced such an effect, as did RM of a grating. However, the adaptation produced by RM was stronger by about 20%. In a second experiment, Von Grunau showed that perceived long-range motion is necessary for adaptation to occur. Anstis, Giaschi, and Cogan (1985) conducted a number of experiments on adaptation to AM. In one, AM was produced by a small spot that oscillated between two positions 1 degree apart with a frequency of 2.5 Hz. For RM the same spot moved sinusoidally across the 1-degree distance with a frequency of 2.5 Hz. Anstis et al. measured the frequency of seeing motion in one stimulus as a function of adaptation to the other. Like von Grunau (1986), Anstis et al. found that adaptation to RM significantly decreased the probability of subsequently seeing AM with the discrete stimuli. However, Anstis et al. found that adaptation to AM produced little or no effect on the probability of seeing RM with the sinusoidally oscillating stimulus. Because of the spatial separations involved, the preceding experiments studied only the interaction between long-range AM and RM and were unable to discriminate between possible effects produced by long-range and short-range AM. However, Gregory and Harris (1984) designed an RM/AM nulling technique that apparently does discriminate between the two AM processes. A sectored disk is mechanically rotated at a constant rate under dim illumination, producing RM in a given direction. Simultaneously, a stroboscope is used to illuminate the disk at a rate that produces AM in the opposite direction. Depending on the number of sectors, rotation rate, and stroboscopic flash rate, the opposite-direction AM can be of only a few arc minutes or over several degrees. The main finding of Gregory and Harris was that for a variety of conditions there exists a luminance intensity ratio for which RM and AM cancel each other out, producing a percept of rapid flicker or random jiggle. However, cancellation occurs only for relatively small jumps (up to about 10 arc min) of the flashed sectors (presumably corresponding to short-range processing). Large jumps (presumably corresponding to long-range processing) do not produce cancellation; rather, two oppositely rotating disks are seen simultaneously. The RM and AM appear not to interact when longrange AM is involved. Most of these studies suggest that long-range AM interacts, at least weakly, with RM. In addition, Gregory and Harris (1984) confirmed the suggestion that short-range AM interacts strongly with RM (see Table 1). On the basis of the studies re- viewed here, one may postulate that both short-range AM and RM are adequate stimuli for low-level neural motion detectors and that long-range AM provides only a weak stimulus for such detectors. This difference in the ability to stimulate neural motion detectors may explain why Gregory and Harris failed to find a nulling effect involving long-range AM and why Barbur (1981) failed to find summation of long-range AM and RM when they moved in opposite directions. The weak stimulation of long-range AM may only be observable when AM and RM are in the same direction and the same low-level motion detectors are activated. Because low-level motion detectors do not generally respond to motion in opposite directions (see Sekuler, Pantle, & Levinson, 1978), a somewhat higher stage of processing is required to observe interactions between motions in opposite directions; this higher order stage may not be activated by the relatively weak stimulation provided by long-range AM. Under any circumstance, if it is true that both AM processes activate low-level motion detectors, then the theoretical proposition that short-range motion occurs at a lower level than longrange movement will require some clarification. Spatial-Frequency Selectivity of Long-Range Process It has been shown in a previous section that <?„,„ for the shortrange process varies inversely with spatial frequency. To complement those studies on the spatial-frequency dependency of the short-range process, this section examines evidence regarding the influence of spatial frequency on long-range AM. Ramachandran, Ginsburg, and Anstis (1983) concluded that low spatial frequencies dominate (long-range) AM on the basis of displays using a competition paradigm: Whenever an initial AM stimulus was followed by two equidistant stimuli, one of which matched some luminance-domain feature of the initial stimulus (e.g., orientation) while the other matched the lowspatial-frequency content of the initial stimulus, AM was always seen in the direction of the low-spatial-frequency stimulus. Ramachandran et al.'s conclusion was supported by the work of von Grunau (1978b), who showed that defocusing stimuli (leaving primarily only low spatial frequencies) in a long-range AM paradigm increased the probability of seeing AM. However, Ramachandran et al. allowed that low spatial frequencies are not necessarily used alone in all occurrences of AM. To demonstrate this point, Prazdny (1986b) replicated the experiments of Ramachandran et al. (1983), but with stimuli that did not differ in spatial-frequency content. Prazdny recognized that, in addition to spatial-frequency descriptions, the initial stimulus in each of Ramachandran et al.'s experiments could be described as a solid figure, whereas the two stimuli in the second frames could always be described as solid figures (corresponding to low-spatial-frequency filtered stimuli) and outline figures (corresponding to high-frequency filtered stimuli). Consequently, he created the same stimuli using dynamic random-dot movies. Solid figures and figure outlines were created by static random dots against a background of dynamic (changing) random dots. In such displays, figures are not denned by intensity changes but by areas in which dots are correlated over time. Like Ramachandran et al., Prazdny found that AM always occurred between solid figures. Unlike the interpretation of Ramachandran et al., Prazdny's results cannot be based on spatial frequency but rather on higher order percep- TWO-PROCESS DISTINCTION IN APPARENT MOTION tual notions of solidness (see the section Relationship Between Short- and Long-Range Motion). to unambiguous horizontal or vertical AM. It was found that The conclusion of Ramachandran et al. (1983) is further frequency of the adaptation and test stimuli were similar or complicated by the results of Watson (1986). In Watson's experiment, two neighboring grating patches abruptly exchanged po- identical (between 1.8 and 3.2 Hz). Further experiments by sitions. It was found that reports of AM only occurred reliably when the spatial frequencies of the two stimuli differed by a factor of 4 or more (over a range of 0-16 cycles/degree). Watson concluded that AM only occurs between similar spatial fre- This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. 119 direction-specific adaptation was strongest when the temporal Kruse et al. supported the notion of frequency-specific processing, and Kruse et al. estimated the bandwidths of the analyzers underlying AM to be 10%. The data of Kruse et al. (1986) are important in implying quencies, but it is clear that the conclusion is stronger than the the existence of temporal-frequency channels that underlie the perception of long-range AM. However, their generalizability is data permit: Watson's experimental design never allowed for limited by the fact that only a narrow range of frequencies was the possibility of seeing AM between dissimilar spatial fre- tested. Furthermore, in their study of adaptation in more con- quencies. Although a number of RM phenomena are spatial-frequency temporal-frequency tuning. Finally, it is questionable whether selective (e.g., motion aftereffects), the data of AM are insufficient or inadequate to make similar claims. Prazdny's (1986b) temporal frequency alone can account for changes in the perception of AM. For example, Anstis et al. (1985) with conven- ventional AM stimuli, Anstis et al. (1985) found no evidence of study showed that low spatial frequencies are not required to tional AM stimuli, Baker and Braddick (1985b) with random- obtain figural selectivity in AM. Watson's (1986) attempt to dot displays, and Petersik and Pantle (1979) with the Ternus show that AM only occurs between similar spatial frequencies suffered from the absence of the most important control condi- differential influences on perception of AM displays. Thus, the tion, one in which AM might have been found between dissimilar spatial frequencies. Therefore, the strongest claim that can same temporal frequency can produce different perceptual effects depending on the relative durations of stimuli and ISIs. be made at this time is that spatial frequency exerts a direct influence on the responses of both the short-range and long- (1938) stimuli found that stimulus duration and ISI have Correspondence Rules for Long-Range AM range processes, although the nature of this influence is not known for the long-range process. Several of the theoretical perspectives discussed in the section Theoretical Points of View suggest that the long-range process Temporal-Frequency Selectivity of Long-Range Process operates at a higher level of perceptual processing than the tionship between the spatial and temporal frequency content of short-range process. This in turn suggests that the long-range process may exhibit a number of flexible heuristics or correspondence rules. Some possibilities are considered here. AM stimuli and the corresponding response properties of the visual system (e.g., Caelli, 1981; Caelli & Finlay, 1979, 1981; Morgan, 1979, 1980a, 1980b; Watson & Ahumada, 1985). adigm in which a single stimulus element in Frame 1 is replaced by two displaced stimuli in Frame 2. One of the second ele- A number of researchers have recently addressed the rela- Common to many of these approaches is the notion that through Fourier analysis, the AM step function can be de- Influence of context. Dawson( 1987) used a competition par- ments (the standard) was a fixed distance from the initial ele- scribed as the sum of a number of low-to-high temporal fre- ment, and the distance of the other was variable. To these stimuli, a variable number of background (context) elements in un- quencies. Because it is the higher temporal frequencies that signal the abrupt transition from stimulus to ISI to stimulus, if the ambiguous motion were added. Dawson measured the distance of the variable second-frame element at which the probabilities visual system were to filter them out, the remaining low frequencies would mimic the stimulus for real oscillatory motion. of seeing AM toward the variable or the standard were equal (a The visual system's response to the low temporal frequencies would then account for the perceived similarity between AM and RM. Dissatisfied with the assumption of a single low-pass filter, (Cruse, Stadler, and Wehner (1986) proposed that AM may be analyzed by a set of frequency-selective RM analyzers. According to this model, analyzers tuned to a given frequency are point of subjective equality [PSE]). Dawson found that context motion in the direction of the standard decreased the PSE of the variable element (in relation to a control condition), apparently because the variable element needed to be nearer the initial element to overcome the influence of context. On the other hand, motion of the context in the direction of the variable element increased its PSE. activated by AM stimuli alternating at a given rate; stimuli Because the spatiotemporal conditions of Dawson's (1987) alternating at different rates would be processed by different displays all favored the long-range process, the results can be interpreted as evidence of a cooperativity effect in long-range analyzers. Kruse et al. (1986) tested this model in a number of experiments on AM adaptation. As a test stimulus, a display was used that is ordinarily bistable: Two dots lying on an imaginary diagonal of a square were alternated with two dots on the other diagonal. With no adaptation, this display yields percepts of either vertical AM or horizontal AM. Under prolonged viewing, these percepts alternate, presumably because of fatigue in the direction-selective analyzers that underlie them. Kruse et al. were able to bias perception of the test display (toward a predominance of either vertical or horizontal AM) by prior adaptation AM. Ramachandran and Anstis (1983) and Ramachandran and Cavanagh (1987) reported a "motion capture" effect that also suggests cooperativity of the long-range process. They showed that when the low spatial-frequency grating was superimposed on two fields of uncorrelated random-dots (which ordinarily would result in the perception of incoherent, random movement), observers perceived the dots as moving coherently, in step with the motion of the grating. Because the long-range motion capture dominated random short-range signals, Ramachandran and colleagues proposed the following solution to the This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. 120 J. TIMOTHY PETERSIK correspondence problem: (a) Motion is initially extracted separately from different spatial-frequency bands; (b) for long-range distances, motion signals from low-frequency channels take precedence over signals from finer image features; (c) fine image features are captured by the low-spatial-frequency motion. Such a model is appealing because it does not require the postulation of higher order processes (like most of the perspectives identified in the section Theoretical Points of View) or complex computations (as in the models of Ullman, 1979, and Hildreth, 1983). The motion capture context effect can also be interpreted as an example of a figure/ground relationship, with low spatial frequencies providing the ground against which the motion of the dots is interpreted. Other examples of figure/ground effects in AM include the work of Ramachandran and Anstis (1986), which showed that square arrangements of dots can be seen in AM when they alternate on a homogeneous background but not when they are embedded in a background array of similar dots. In the latter case, the targets are merely interpreted as part of a nonmoving background. Ramachandran and Anstis also showed that gaps in textured background only participate in AM when they are highlighted by outlines. Petersik and McDill (1981) showed that patterns of erasure in an otherwise static set of background elements can create the impression of a moving figure. In their display a complete set of regularly spaced white bars was initially presented on a black background. Starting on one side of the pattern, bars were successively erased and replaced, the reappearance of one bar in the pattern coinciding with the erasure of the next. This sequence of events mimicked the successive optical occlusion of a background pattern by a moving object, and the form and motion of that object were subsequently constructed in the percepts of the subjects.7 Anstis et al. (1985) demonstrated an AM adaptation effect that depended on context. Ordinarily, adaptation to an AM display consisting of two horizontally separated dots results in a changeover from the experience of motion to the experience of flicker or alternation without motion. In one of their experiments, Anstis et al. added two context elements to a display of two such horizontally separated elements. The addition of the contextual elements left the two critical stimuli alternating in their original spatiotemporal relationship, but the experience of AM changed from horizontal to vertical; that is, the AM signal that originally was between the two critical stimuli now changed so that AM was from each critical stimulus to a context element. After prolonged viewing of this four-dot display, subjects failed to show an adaptation effect when presented with a two-dot horizontal AM display; removing the context elements restored the adaptation effect. Anstis et al. concluded that adaptation is produced by perceived motion, not by the mere fact of stimulus alternation. Ramachandfan, Inada, and Kiama (1986) have reported a case in which a moving context seems to facilitate, if not produce, the percept of movement of a single, unpaired stimulus dot. In the first frame of an AM display, eight dots were plotted on a cathode ray tube screen. In the second frame, all dots were displaced a constant distance and direction, but a piece of opaque tape occluded the appearance of one of them. Nonetheless, when the frames alternated, observers reported seeing the motion of the dot whose partner was occluded. The dot gave the appearance of moving behind the piece of tape attached to the screen. This display again suggests a cooperativity effect in longrange AM. However, little background motion is necessary to produce motion of the single dot: A context consisting of one unambiguously moving dot was almost as effective as a context consisting of four. Whereas Ramachandran et al. showed how motion in the background can facilitate the perception of motion of a stationary dot, Banta and Breitmeyer (1985) reported the opposite: Stationary flanking bars in an AM display can inhibit the perception of motion between two alternating target bars. Another cooperative effect that depends on context was described by Anstis and Ramachandran (1986). In their display, four out of five stimulus dots each traveled a V-shaped path in three frames of AM. The fifth (central) stimulus dot was not presented at the corner of the V in the second frame, instead reappearing at the end location in the third frame. Although the path defined by the fifth dot was horizontal, and would ordinarily be perceived as such if presented on a contextless background, the presence of the background dots moving along their V-shaped paths caused the fifth dot to similarly appear to travel a V-shaped path. The results of experiments examining the influence of context suggest a number of conclusions. It seems incontrovertible that the motion or stationarity of context elements influences the nature of the motion of any test element. This in turn indicates a cooperativity of local long-range motion-analyzing mechanisms; at some stage of processing there must be a network of excitatory or inhibitory connections between directionselective and velocity-selective elements. Also, it is clear that the higher order long-range process operates by identifying objects (figures) against a background. This suggests that the long-range process may be making a "best guess" about the nature of events in the world that produce the proximal pattern of stimulation (Rock, 1983; but see the section Are correspondence rules "perceptually intelligent"?, in this article). Finally, it is apparent that for the long-range process, adaptation effects are critically dependent on perceived motion, not merely retinal stimulation, and that perceived motion is dependent on its context. Form versus motion information. The preceding paragraph suggests that the long-range process relies on the identification of forms to which the motion sensation can be imparted. In fact, the correspondence problem itself only makes sense if the assumption is made that there are forms (however small or local) that need to be matched over time. Nonetheless, there has always been some belief that motion percepts can exist in a pure state in the absence of forms to carry them. The notion of Ramachandran et al. (1983) ^hat low spatial frequencies dominate AM supports this conjecture inasmuch as low frequencies do not provide detail for form perception. Two experiments by von Grunau (1978a, 1979) addressed the relationship between form and AM perception. Von Grunau used metacontrast masking to inhibit the form information present in the first or second stimulus or both in a long-range AM display. The results were that masking both AM stimuli nearly eliminated reports of AM 7 The display of Petersik and McDill (1981) was also bistable: The moving subjective figure was only seen when the rate of erasure/replacement was relatively fast. For slower rates, subjects perceived individual motions of the bars. At intermediate rates, the percepts alternated. It is not known whether these bistable percepts correspond to short-range and long-range phenomena. TWO-PROCESS DISTINCTION IN APPARENT MOTION (von Grunau, 1979); masking the first stimulus only reduced, the 2-D plane of the stimuli or in 3-D space, in which case it but did not eliminate. AM; and masking the second stimulus would appear much like a page turning in a book. Kolers and only increased the likelihood of seeing AM (von Grunau, Green varied ISI as the parameter (0-501 ms) and found that 1978a). Together, these experiments show that form informa- complex rotations of the AM stimulus (usually three-dimen- tion, especially from the first stimulus of an AM pair, is neces- sional) occurred only at the longer ISIs. Furthermore, when 3D rotation did occur, it was equally likely to be seen with stimuli sary for the long-range process to generate a motion signal. The relationship between form and motion is considered again in the section Relationship of Form and Motion. Three-dimensional spatial separations- Ullman (1979) suggested that the correspondence process is solved on the basis of two-dimensional (2-D), or retinal, separations, not on the basis of perceived three-dimensional (3-D) separations. This was This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. 121 based on his finding (Ullman, 1978) that when lines participating in AM were presented within the context of a perspective for which there was no color disparity. The interpretation of this finding is that color disparity itself is not a condition that promotes 3-D rotation. Over a large range of ISIs, the AM process preferred a correspondence rule that yielded a direct path between the rectangular shapes. The similarity between the appearances of stimuli that did and did not contain color disparities suggests that AM was based on rule-based processing, not perceptual intelligence. drawing (that suggested various 3-D distances), it was the 2-D A similar conclusion can be drawn from the results of Peter- distance only that mediated correspondence matching. A similar conclusion was reached by Mutch, Smith, and Yonas (1983) sik (1987). He modified the Ternus (1938) stimuli in such a way that, in the absence of alternation, the dotlike elements in both when they varied the actual 2-D and 3-D distances of AM stim- frames gave rise to the appearance of a subjective figure. Alternation of the frames under conditions favorable to the long- uli. However, Green and Odom (1986) have described a display in which AM paths are determined by 3-D proximity. Using stimuli that were displayed in AM stereoscopically, Green and Odom were able to pit retinal correspondences against correspondences existing in different depth planes. They found that correspondence matches were always biased toward neighbors in the same depth plane and that the degree of bias was a positive function of the degree of stereoscopic depth. A control ex- range process always led to perception of the subjective figure. However, alternation of the frames under conditions favorable to the short-range process either eliminated the appearance of the subjective figure or gave rise to a form of AM that was qualitatively different from that experienced under long-range conditions. Petersik's conclusion was that AM of the subjective figure periment showed that the original data were produced by the depended on the perceptual fate of the inducing elements, which was in turn determined by the kind of AM that the ele- long-range process. The finding of Green and Odom (1986) is not necessarily in- Green (1984), Petersik argued that a perceptually intelligent so- compatible with Ullman's (1979) hypothesis or data (Ullman, 1978), or with the data of Mutch et al. (1983). In the earlier studies, the first AM stimulus was always presented in a given depth plane, whereas second (competing) stimuli were in the same and different planes. Given Green and Odom's discovery ments gave rise to. Reasoning in the manner of Kolers and lution in both cases would have been to impart smooth AM to the subjective figure. Implications for Theory that the long-range process prefers matches made in the same depth plane, it is not surprising that the earlier studies failed to The lack of reliable findings concerning the spatial and temporal selectivity of the long-range process, coupled with its ap- find AM across perceived depth planes. The best interpretation parent flexibility with regard to context, make it difficult to associate the long-range process with the more "hard-wired" pro- of all studies is that the long-range system prefers correspondence matches within the same depth plane, or perhaps neigh- cesses discovered in physiological experiments and in relatively boring depth planes, and avoids matches across depth planes. This seems to be a useful correspondence rule inasmuch as real- constrained psychophysical experiments (e.g., transient channels or cortical motion detectors). Similarly, it is difficult to see world motions rarely involve local jumps of single elements across depth planes. When motion does produce changes in depth, they are generally gradual and involve collections of ele- how the properties of the long-range process might be explained by what is presently known about influences on visual persistence. On the other hand, the existence of multiple context ments, conditions that might favor processing by the shortrange process. Are correspondence rales "perceptually intelligent1".' A num- effects and form segregation located in depth suggest a flexible higher order kind of processing, although the long-range process ber of authors (e.g.. Rock, 1983) have characterized perception detectors as to imply a perceptually intelligent, unconscious in- as a process of, perhaps unconsciously, deriving intelligent solu- ference at work. When considered alongside those of the shortrange process, the properties of the long-range process suggest tions to problems posed by stimulus configurations. Kolers and Green (1984) challenged that approach as it applies to longrange AM. Their experimental stimuli consisted of alternating rectangular shapes separated by 2.4 arc degrees. Each rectangle comprised two smaller boxes that were colored either red or green. Thus, for example, one rectangle might consist of (from left to right) green and red squares and its displaced partner consist of red and green squares. Kolers and Green argued that an intelligent way to resolve the change in the positions of the colored components of the stimuli would be to perceptually rotate the stimulus during AM. Such rotation could take place in may not be so free from the constraints of underlying motion that the most fruitful theoretical perspective may be one emphasizing peripheral motion detectors (short range) and more central perceptual processing (long range). Issues Involving Short- and Long-Range Processes Relationship of Form and Motion The results of Dick et al. (1987; see also the section Discriminating Functionally Different Perceptual Processes, in this arti- This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. 122 J. TIMOTHY PETERSIK cle) led to the conclusion that long-range motion may depend on the conjunction of the detection of the appearance and disappearance of stimulus forms, whereas motion may be detected more directly (by a preattentive process) when it is short range. Nonetheless, the question arises regarding the extent to which the motion signals generated by the short-range process depend on form information. Work with the Ternus (1938) display implies that the short-range process relies on figural similarity between neighboring elements. Using vertical line stimuli rather than dots, Pantle and Petersik (1980) found that orientation perturbations of the overlapping elements of as little as 3° decreased the frequency of reports of end-to-end movement. This could be interpreted to mean that the short-range process, which presumably signals the stationarity of the overlapping elements in the Ternus display, is sensitive to form differences in the stimuli. The fact that group movement can be perceived with such orientation perturbations may mean that the longrange process is insensitive to form differences (see also Kolers & von Grunau, 1976). Petersik (1984) alternated two different stimuli that had the same spatial arrangement as in the normal Ternus display but that differed in their figural properties (Stimulus 1 consisted of the letters ITEM; Stimulus 2 consisted of the letters AVHW shifted horizontally by one letter width). Even with stimulus alternation rates that favored the short-range process, subjects failed to see anything corresponding to end-toend movement. Rather, subjects ignored form information and perceived motion of a group of elements. Thus, at least with the Ternus display, changes in the form of stimulus elements across frames reduce the frequency of percepts assumed to correspond to the short-range process. In contrast to the short-range process, details of form seem to be irrelevant to the long-range process. As has been reported elsewhere (e.g., Kolers & von Grunau, 1976; Navon, 1976), long-range motion is apparently insensitive to changes having to do with the shape and other details of the inducing stimuli. Group movement can be generated equally well when the stimuli between frames are very different as when they are identical; however; slight differences in local form can disrupt the shortrange process (Pantle & Petersik, 1980; Petersik, 1984). Another characteristic of the long-range process is the tendency to group elements or join them into functional units, especially when there is no necessary point-by-point correlation between frames (Petersik & Pantle, 1979). This is in opposition to the short-range process that groups elements only on the basis of point-by-point similarity. Furthermore, the grouping produced by the long-range process seems to occur on a larger scale, or with the larger tokens (perhaps synthesized from smaller stimulus elements), than do those produced by the short-range process. These proposed differences in sensitivity to figural detail between the short- and long-range processes correspond well to observations made by others, particularly those having to do with the suggestion that the short-range process is more selective for high spatial frequencies, whereas the long-range process is more selective for low spatial frequencies (cf. Breitmeyer, Love, & Wepman, 1974; Breitmeyer & Ritter, 1986a, 1986b; Petersik & Pantle, 1979;Ramachandranetal., 1983). Because of its sensitivity to spatial detail and contrast (as well as to the timing of stimulation), another way to describe the short-range process is that it exhibits lability; that is, slight stimulus manipulations can disrupt its activity. On the other hand, the long-range process's insensitivity to spatial detail suggests tolerance. It is more concerned with the general size and location of higher order clusters than with details of form. Relationship Between Short- and Long-Range Motion As was noted earlier, Julesz (Chang & Julesz, 1983a) stated that his local, feature-based mechanism achieves some preprocessing prior to a global correlation. On the other hand, both Braddick (1980) and Anstis (1980) interpreted the short-range process (which corresponds to Julesz's global process) as being low-level and therefore preceding the higher order long-range process. Studies by Petersik, Hicks, and Pantle (1978) and by Prazdny (1986a) showed that the output of the short-range process can serve as an input to the long-range process. For example, Petersik et al. relied on the short-range process to produce percepts of bar stimuli in two alternating random-dot frames. Then, the original two random-dot frames were replaced by a second pair of random-dot frames whose alternation produced percepts of bar stimuli in new spatial locations. The change in spatial location was chosen to mimic the arrangement of physical contours in the Ternus (1938) display. The stimulus bars were defined either as correlated regions (over time) against an uncorrelated background or as uncorrelated regions against a correlated background. Additionally, when the bars were defined by correlation, the central overlapping bar could maintain its correlation across pairs of random-dot frames or change its correlation across them. Petersik et al. found that percepts corresponding to group and end-to-end movements could be obtained with such displays. End-to-end-like movement was most reliably obtained when the correlation of the overlapping bar was maintained across both pairs of alternating random-dot frames. Grouplike movement was obtained when there was no spatial correlation between the elements of the overlapping bars across pairs of random-dot frames. To account for the data, Petersik et al. proposed a two-stage model in which the figures defined (across any pair of random-dot frames) by the shortrange process feed into a higher order AM process. The higher order stage, in turn, generates signals of global AM involving the figures defined by the short-range process. Nonetheless, when a correlation persists in a single location over different pairs of random-dot frames, the short-range signal for stationarity dominates, and that figure does not participate in global AM generated by the higher order process. Prazdny (1986a) drew similar conclusions. In his experiment, subjects were able to "effortlessly and spontaneously" perceive kinetic depth effects with stimuli that were defined solely by the random (shortrange) movement of target dots against a stationary background of random dots. Whereas the preceding experiments show that the shortrange process can feed the long-range process, they fail to demonstrate that such stages are necessary, that is, that the shortrange process necessarily feeds the long-range process. In fact, the observation that the outside, unpaired, dots participate in long-range AM at the same time that the overlapping dots are perceived as stationary (or jiggling, when there is a small perturbation) in end-to-end movement of the Ternus (1938) display implies that the long-range and short-range processes can be simultaneously activated. The two processes can also be made rivalrous (e.g., Pantle & Picciano, 1976; see also the section His- This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. TWO-PROCESS DISTINCTION IN APPARENT MOTION torical Development of Concept of Two AM Processes, in this article). Mather, Cavanagh, and Anstis (1985) observed a rivalry with a display consisting of a modified square-wave grating in AM. Every 10th location in a sequence of black and white bars was occupied by a bar of double width. On each frame of the display, the double-width bar was shifted in one direction (by a single bar width) and simultaneously changed its polarity (i.e., changed from black to white, or vice versa). Because the doublewidth bar formed a figure on a background of smaller bars, and because the long-range process is insensitive to contrast (Pantle & Picciano, 1976), activity in the long-range process would lead to global motion of the double-width bar in the direction of the shift. However, because the short-range process is sensitive to contrast, it would detect motion of like-polarity edges in the direction opposite that of the shift of the double-width bar. At a fast enough rate of presentation, the observer should have a sense of relatively continuous motion in one direction if the long-range process is activated, and a sense of a number of discrete motions in the opposite direction if the short-range process is activated. For frame durations of 16.7 ms and no ISIs, Mather et al. observed that the display was rivalrous; only one direction of motion was seen at any time. The direction of motion actually observed depended to a great extent on retinal location and bar width. Peripheral viewing favored long-range motion. Central viewing favored short-range motion for small bar widths (2 arc min) and long-range motion for larger widths (8 arc min). In support of Braddick's (1974) suggestion that the short-range process is based on the output of directionally selective neural motion detectors, Mather et al. found that their display always generated motion aftereffects in the direction opposite the short-range movement component, irrespective of the direction of motion reported by the observer. Finlay, Manning, Neill, and Fenelon (1987) showed that the presence of short-range motion in the background of a display can result in an upward shift in the optimal frequency for producing long-range AM. As they noted, their data suggest that the short- and long-range processes interact or perhaps share common processing mechanisms. The available evidence suggests that the short- and long-range processes, if they are indeed distinct perceptual mechanisms, do not constitute necessary sequential stages of an AM-generating process: Each seems capable of creating coherent percepts on its own. Furthermore, when present in the same display, their activities may interact. As was noted by Prazdny (1986a), they are most likely cooperative, or complementary, processes whose dissociation is only observed in specialized laboratory situations. Failures to Find Two Processes The evidence reviewed thus far has focused on reports favorable to the two-process interpretation of AM. However, a number of researchers have failed to find any distinction between results obtained under conditions allegedly favorable to the short- and long-range processes, respectively. For example, in an experiment by Mutch, Smith, and Yonas (1983), a single center stimulus was followed by two (competing) AM stimuli, one each to the right and left of center. The second stimuli varied in their 2-D and 3-D distances from the initial stimulus. Mutch et al. 123 found that motion of the center dot to one of the side dots, as well as motion quality ratings, were based on 2-D distances alone and that the data could be fit well by straight lines, regardless of the fact that the stimulus separations were varied over distances preferred by both the short- and long-range processes. Kalbaugh and Petersik (1989) studied the spatiotemporal parameters of a visual interpolation effect in AM that was initially reported by Shaw and Ramachandran (1982). In this display, AM was produced by the sequential presentation of vertical columns of dots, each column displaced by a constant amount in relation to the previous column. When one of the middle columns is left blank (not plotted), the gap is not detected under certain spatiotemporal conditions; that is, AM is seen through the gap. Kalbaugh and Petersik varied the spatial separation of the columns along with the stimulus duration. Their dependent measure was the ISI at which the gap became detectable. Like Mutch et al. (1983), Kalbaugh and Petersik found that their data could be described by a continuous straight line: The critical ISI fell as stimulus duration increased and was independent of the spatial separation of the columns. There was no empirical or subjective evidence of a difference between short- and longrange processes, despite the fact that different spatiotemporal conditions favored both. Some of the studies just reviewed that have addressed the limiting conditions of the short-range process have failed, in the end, to discriminate between the short- and long-range processes. For example, in their study of the segregation of displaced and correlated areas in equiluminous random-dot displays, Cavanagh et al. (1985) reported that a single motion process seemed to account for the perception of displays over a four-octave range in element size and for displacements that reached 2 degrees. Furthermore, they reported that the conditions needed to perceive coherent motion in random-dot displays did not differ from those required to perceive motion with an isolated bar stimulus. Therefore, they concluded that no aspects of their data seemed to warrant the assumption of two separate AM processes. Similarly, in their study of the indistinguishability of AM and RM with periodic stimuli, Burr et al. (1986) found that their results gave "no indication of a dichotomy which may parallel the putative short and long range processes" (p. 651). Perhaps the most comprehensive study suggesting a unitary AM process is that of Burt and Sperling (1981). In their displays, a row of dots in one frame is replaced after a time interval, t, by a second row whose elements have been displaced by horizontal and vertical distances (and so on over several frames). Although there are several alternative motion paths, subjects typically see motion in a single direction at any moment. Burt and Sperling found that for different horizontal and vertical displacements, the value of t needed to produce equilibrium between any pair of alternative motion paths was a log linear function of the alternate distances (</,•, dj) from a dot in the first frame to its potential partner in the second. In general, their results show no need to postulate two independent mechanisms mediating AM. In a related study, Falzett and Lappin (1983) concluded that there is no need to postulate a short-range process with a spatial limit of 15 arc min to account for the detectability of stationary and moving targets when target and background elements are distributed in time and space. The failure to find a distinction based on the short- and long- 124 J. TIMOTHY PETERSIK range processes in these displays emphasizes the lack of ability to predict the precise circumstances under which the two pro- COMMON MOTION REPRESENTATION cesses will manifest themselves as well as the lack of a theoreti- /T\ cal framework with which to interpret them. Conclusion LONG-RANGE MOTION Heuristic Theoretical Perspective PROCESSING /TN The studies reviewed in this article have generally satisfied the This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. first six of the criteria established in the section Discriminating Functionally Different Perceptual Processes, regarding when different percepts can be attributed to functionally different perceptual processes. It is the purpose of this section to take SHORT-RANGE SPATIAL POOLING a first step toward satisfying the last criterion, namely, toward establishing a theoretical framework for understanding the operation of the two processes. It is well established that the visual system has effected an equilibrium between the need for sensi- SHORT-RANGE MOTION DETECTION FORM PROCESSING tivity to light over a broad range of luminances and the need for sensitivity to stimulus detail by evolving a duplex retina. The cone system (and related pathways) achieves sensitivity to detail at the expense of sensitivity to light, whereas the rod system (and related pathways) achieves sensitivity to light at the expense of sensitivity to detail. Similarly, it may be postulated that the visual system has effected a balance between sensitivities to motion and various stimulus details by evolving two motion- DYNAMIC LUMINANCE DISTRIBUTION Figure 3. Schematic model of the processes involved in motion perception based on the findings of this review. processing systems. The short-range system achieves point-bypoint correlations based on stimulus luminance at the expense of being limited to a small spatial scale (perhaps determined by the stimulus grain size, as suggested by Chang & Julesz, 1983h, stages). A dynamic luminance distribution gives rise to both short-range motion detection, based on the outputs of low-level and by Petersik et al., 1983). It may function at two levels, one directionally selective motion detectors, and form processing (if defining contours or features are present). The outputs of the reflecting the output of simple motion detectors, and the other reflecting a spatial pooling of such detectors. In so doing, it can segregate forms, as in random-dot displays. Because it is a relatively low level operation and is largely stimulus driven, the motion detectors are integrated at the level of the short-range spatial pooling process. It is in this process that the generation short-range system is labile; that is, slight perturbations in spatial, temporal, or contrast factors can inhibit its functioning. of figures based on local motions and global cooperativity effects occurs. The pooling process, in turn, can feed the longrange motion-detecting process, as can the output of form pro- Similarly, it is influenced by relatively peripheral physiological cessing. The long-range process itself may rely on low-level mo- effects, for example, visual persistence. On the other hand, the long-range process achieves stimulus matches over large dis- tion to a certain extent, but its functioning is primarily thought to be of a higher level, heuristic nature. Finally, the outputs of tances, but at the expense of being insensitive to stimulus details. As such, the long-range process is susceptible to context each of the motion-detecting processes enter a common motion representation, at which an interpretation is made on the basis effects and various top-down influences. In general, it can be of the strengths and nature of the inputs. The model of Figure 3 can be used to explain the percepts thought of as flexible. It operates over a wide range of spatial, temporal, and contrast parameters and may function to provide obtained with a variety of displays. For example, the incoherent an object-based interpretation of stimulus displacements (cf. motion perceived with the alteration of uncorrelated random- Ramachandran & Anstis, 1986; Ramachandran et al., 1986). dot displays is thought to reflect the direct output of low-level Whereas the short-range process is capable of generating stimu- motion detectors (short-range motion detection). No correla- lus forms, the long-range process may require stimulus forms tion exists to stimulate global cooperative processing (spatial pooling), and there are no stimulus forms available to feed the long-range process. On the other hand, when the alternating for its input (Anstis, 1978). Together, the short- and long-range processes can thus tolerate a broad range of spatial displacements while maintaining good sensitivity to stimulus details. As Prazdny (1986a) suggested, the two processes probably ordinarily operate in a complementary fashion. With some displays the short- and long-range processes can in fact be brought into competition, suggesting that they share at least some processing resources. Figure 3 is a schematic attempt to represent some of the preceding notions. In this model, motion perception is conceived of as involving a series of staggered processes (as opposed to random-dot frames include correlated and displaced patches, the outputs of low-level motion detectors can be pooled to generate a stimulus form, which is identified and seen as "jiggling." In this case, because the correlated patch is not displaced a great distance, no long-range processing enters. However, if the correlated patch itself is shifted sufficiently, the outputs of the spatialpooling process can enter the long-range process, and the patch will itself be seen in AM. If the dynamic luminance display itself consists of readily identifiable stimulus forms, the model of Fig- This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. TWO-PROCESS DISTINCTION IN APPARENT MOTION ure 3 suggests that the output from the level of form processing stimulates the long-range AM process directly, as in classical AM. The percept that dominates at the common motion representation is thought to depend on, at minimum, how well the spatial and temporal characteristics activate the short- or longrange processes or both and on how well the motion interpretation (or interpretations) matches one's internal model of the environment, that is, how ecologically justified the interpretation is. Take, for example, a display used by Pantle and Petersik (1980) in which the Ternus (1938) dots of Figure 1 were replaced by vertical line segments. In addition, the overlapping lines of one frame were tilted by 6 degrees around their midpoints. According to the model of Figure 3, if the alternation rate stimulates short-range motion detectors, the overlapping lines should be seen as shifting about their centers in rotational AM. This is in fact seen. However, the outer lines, whose separation is too great to be registered by simple motion detectors, also activate the long-range process (through form processing) and are seen in AM. On the other hand, if the alternation rate is so long that motion detectors cannot be activated, the longrange process groups the collection of lines in each frame, and the observer perceives the entire collection of elements shifting back and forth. The orientation shift of the overlapping lines is hardly noticeable. The advantage to a model like that shown in Figure 3 is that the individual components can themselves be modeled and updated as new information is made available. Thus, for example, either or both of the short-range processes can be interpreted in terms of visual persistence phenomena or in terms of sustained channels. At the moment, the early short-range process is modeled by low-level motion detectors. The validity of this assumption, along with the relationship between the two AM processes and RM, is examined in the following section. 125 perception of motion to objects whose proximal forms change while they move. Such a case might occur, for example, when an animal springs from a relatively compact position in which its limbs and body are pulled together to a position in which its limbs and body are extended. Because the bodies of animals also tend to be textured, one might expect that the perception of such biological motion ordinarily relies on a cooperation between local short-range analysis of texture and bodily parts (Johansson, 1975) and global long-range form perception. One value of considering the short- and long-range processes in this way is that it forces the consideration that they, and AM phenomena in general, exist in the context of perceptual events that themselves exist to permit the observer to interact eifectively with the environment, to validly interpret changes in the proximal stimulus, and to predict future events. In this respect, future analyses of these processes may benefit from a consideration of how they function in the presence of naturally occurring image motion and eye movements (e.g., Post & Leibowitz, 1985; Swanston, Wade, & Day, 1987, especially pp. 153-155). References Anstis. S. M. (1970). Phi movement as a subtraction process. Vision Research, 10, 1411-1430. Anstis, S. M. (1978). Apparent movement. In H.-L. Teuber, R. H. Held, & H. W. Leibowitz (Eds.), Handbook of sensory physiology: Vol. 8. Perception. New York: Springer-Verlag. Anstis, S. M. (1980). The perception of apparent movement. Philosophical Transactions of the Royal Society of London, Series B, 290, 153- 168. Anstis, S. M. (1986). Motion perception in the frontal plane: Sensory aspects. In K.. R. Boff, L. Kaufman, & J. P. Thomas(Eds.), Handbook of perception and human performance (Vol. I): Sensory processes and perception. New York: Wiley. Anstis, S., Giaschi, D.. & Cogan. A. J. (1985). Adaptation to apparent motion. Vision Research, 25. 1051-1062. Anstis, S. M., & Ramachandran, V. S. (1986). Entrained path deflection Relationship to RM in apparent motion. Vision Research, 26, 1733-1739. Baker, C. L., & Braddick, O. J. (1982a). The basis of area and dot num- If the visual system indeed evolved a duplex motion-analyzing system, it is logical to ask what functions the two systems serve in the natural world, where AM does not occur. For a short-range system whose building blocks are motion detectors themselves, the answer seems plain: Short-range AM is simply a laboratory consequence of the fact that cells designed to respond to RM respond equally well to discrete stimuli having the proper spatial and temporal parameters (Orban, 1986). Hence, the short-range process's function may simply be to segregate moving objects on the basis of local motion signals arising from their parts. However, the ecological function of a long-range process seems less apparent. The model presented in Figure 3 suggests two possibilities. Because the long-range process is hypothesized to obtain some input from spatially pooled motion detectors, it is possible that its response to AM is also a consequence of the fact that RM and AM can have similar spatiotemporal representations (Watson & Ahumada, 1985). Thus, in comparison with the short-range process, the long-range process can be hypothesized to use a spatiotemporal filter that favors longer stimulus displacements. At the same time, because the representation of Figure 3 suggests that it also receives inputs from a form-processing level, it can be hypothesized that the long-range process functions to continuously impart the ber effects in random dot motion perception. Vision Research, 22, 1253-1259. Baker. C. L., & Braddick, O. J. (1982b). Does segregation of differently moving areas depend on relative or absolute displacement? Vision Research, 22, 851-856. Baker, C. L., & Braddick. O. J. (1985a). Eccentricity-dependent scaling of the limits for short-range apparent motion perception. Vision Research, 25, 803-812. Baker, C. L., & Braddick, O. J. (1985b). Temporal properties of the short-range process in apparent motion. Perception, 14. 181-192. Banta, A., & Breitmeyer, B. G. (1985). Stationary patterns suppress the perception of stroboscopic motion. Vision Research, 25, 1501-1505. Barbur, J. L. (1981). Subthreshold addition of real and apparent motion. Vision Research, 21, 557-566. Barlow, H. B.. & Levick. W. R. (1965). The mechanism of directionally selective units in rabbit's retina. Journal of Physiology, 178, 447-504. Bell, H. H., & Lappin, J. S. (1979). The detection of rotation in randomdot patterns. Perception & Psychophysics, 26, 415-417, Bischof, W. E, & Groner, M. (1985). Beyond the displacement limit: An analysis of short-range processes in apparent motion. Vision Research, 25, 839-848. Braddick, O. (1973). The masking of apparent motion in random-dot patterns. Vision Research. 11, 355-369. Braddick, O. J. (1974). A short-range process in apparent motion. Vision Research, 14, 519-527. This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. 126 J. TIMOTHY PETERSIK Braddick, O. J. (1980). Low-level and high-level processes in apparent motion. Philosophical Transactions of the Royal Society of London, Series B, 290, 137-151. Braddick, O. J., & Adlard, A. J. (1978). Apparent motion and the motion detector. In J. Armington, J. Krauskopf, & B. R. Woolen (Eds.), Visual psychophysics and physiology (pp. 417-426). New York: Academic Press. Breitmeyer, B. G. (1984). Visual masking: An integrative approach. New York: Oxford University Press. Breitmeyer, B. G., Love, R., & Wepman, B. (1974). Contour suppression during stroboscopic motion and metacontrast. Vision Research, H-1451-1456. Breitmeyer, B. G., & Ritter, A. (1986a). The role of visual pattern persistence in bistable stroboscopic motion. Vision Research, 26, 18011806. Breitmeyer, B. G., & Ritter, A. (1986b). Visual persistence and the effect of eccentric viewing, element size, and frame duration on bistable stroboscopic motion percepts. Perception & Psychophysics, 39, 275280. Burr, D. C, Ross, J., & Morrone, M. C. (1986). Smooth and sampled motion. Vision Research, 26. 643-652. Burt, P., & Sperling, G. (1981). Time, distance, and feature trade-offs in visual apparent motion. Psychological Review, 88, 171-195. Caelli, T. (1981). Visual perception: Theory and practice. New York: Pergamon Press. Caelli, T., & Finlay, D. (1979). Frequency, phase, and colour coding in apparent motion. Perception, 8, 59-68. Caelli, T, & Finlay, D. (1981). Intensity, spatial frequency, and temporal frequency determinants of apparent motion: Korte revisited. Perception, 10, 183-192. Cavanagh, P., Boeglin, J., & Favreau, O. E. (1985). Perception of motion in equiluminous kinematograms. Perception, 14, 151-162. Chang, J. J., & Julesz, B. (1983a). Displacement limits, directional anistropy and direction versus form discrimination in random-dot cinematograms. Vision Research, 23, 639-646. Chang, J. J., & Jules/, B. (I983b). Displacement limits for spatial frequency filtered random-dot cinematograms in apparent motion. Vision Research, 23. 1379-1385. Chang, J. J., & Julesz, B. (1984). Cooperative phenomena in apparent movement perception of random-dot cinematograms. Vision Research, 24. 1781-1788. Chang, J. J., & Julesz, B. (1985). Cooperative and non-cooperative processes of apparent movement of random-dot cinematograms. Spatial Vision, 1, 39-45. Clatworthy, J. L., & Frisby, J. P. (1973). Real and apparent visual movement: Evidence for a unitary mechanism. Perception, 2, 161-164. Dawson, M. R. (1987). Moving contexts do affect the perceived direction of apparent motion in motion competition displays. Vision Research, 27, 799-809. Dick, M., Ullman, S.. & Sagi, D. (1987). Parallel and serial processes in motion detection. Science, 237, 400-402. Falzett, M., & Lappin, J. S. (1983). Detection of visual forms in space and time. Vision Research, 23, 181-189. Finlay, D. C., Manning, M. L., & Fenelon, B. (1987). Individual differences in responses of untrained observers to stroboscopic apparent motion. Perception, 16. 573-582. Finlay, D. C., Manning, M. L., Neill, R. A., & Fenelon, B. (1987). Effects of movement in the background field on long-range apparent motion. Vision Research. 27. 1679-1682. Gerbino, W. (198!). Monoptic and dichoptic signals do not cooperate in the perception of a bistable motion display. Ada Psychologies, 48, 79-87. Gerbino, W. (1984). Low-level and high-level processes in the perceptual organization of three-dimensional apparent motion. Perception, 13, 417-428. Green, M. (1983). Inhibition and facilitation of apparent motion by real motion. Vision Research, 23, 861-865. Green, M., & Odom, 1. V. (1986). Correspondence matching in apparent motion: Evidence for three-dimensional spatial representation. Science. 233, 1427-1429. Gregory, R. L., & Harris, J. P. (1984). Real and apparent movement nulled. Nature, 307, 729-730. Hildreth, E. C. (1983). The measurement of visual motion. Cambridge, MA: MIT Press. Johansson, G. (1975). Visual motion perception. Scientific American, 232, 76-88. Julesz, B. (1971). Foundations of cydopean perception. Chicago: University of Chicago Press. Julesz, B., & Schumer, R. A. (1981). Early visual perception. Annual Review of Psychology. 32. 575-628. Kalbaugh, R., & Petersik, J. T. (1989). Spatiotemporal aspects of an interpolation effect in apparent motion. Unpublished manuscript. Kaufman, L., Cyrulnick, I., Kaplowitz, J., Melnick, G., & Stof, D. (1971). The complementarity of apparent and real motion. Psychologische Forschung, 34, 343-348. Kolers, P. A. (1972). Aspects of motion perception. New York: Pergamon Press. Kolers, P. A., & Green, M. (1984). Color logic of apparent motion. Perception. 13, 249-254. Kolers, P. A., & von Grunau, M. (1976). Shape and color in apparent motion. Vision Research, 16, 329-335. Kruse, P., Stadler, M., & Wehner, T. (1986). Direction and frequency specific processing in the perception of long-range apparent movement. Vision Research, 26, 327-336. Lappin, J. S., & Bell, H. H. (1976). The detection of coherence in moving random-dot patterns. Vision Research, 16, 161-168. Larsen, A., Farrell, J., & Bundeson, C. (1983). Short- and long-range processes in visual apparent movement. Psychological Research, 45, 11-18. Livingstone, M. S., & Hubel, D. H. (1987). Psychophysical evidence for separate channels for the perception of form, color, movement, and depth. Journal of Neuroscience, 7, 3416-3468. Livingstone, M., & Hubel, D. (1988). Segregation of form, color, movement, and depth: Anatomy, physiology, and perception. Science. 240, 740-749. Mather, G., Cavanagh, P., & Anstis, S. M. (1985). A moving display which opposes short-range and long-range signals. Perception, 14, 163-166. Morgan, M. J. (1979). Perception of continuity in stroboscopic motion: A temporal frequency analysis. Vision Research, 19, 491-500. Morgan, M. (1980a). Analogue models of motion perception. Philosophical Transactions of the Royal Society of London, Series B, 290, 117-135. Morgan, M. (1980b). Spatiotemporal filtering and the interpolation effect in apparent motion. Perception, 9, 123-248. Morgan, M. J., & Ward, R. (1980). Conditions for motion flow in dynamic visual noise. Vision Research, 20, 431 -435. Mutch. K., Smith, I. M., & Yonas, A. (1983). The effect of two-dimensional and three-dimensional distance on apparent motion. Perception, 72.305-312. Nakayama, K.. (1985). Biological image motion processing: A review. Vision Research, 25, 625-660. Nakayama, K., & Silverman, G. H. (1984). Temporal and spatial characteristics of the upper displacement limit for motion in random dots. Vision Research, 24, 293-300. Navon, D. (1976). Irrelevance of figural identity for resolving ambiguities in apparent motion. Journal of Experimental Psychology: Human Perception and Performance. 2, 130-138. Orban, G. A. (1986). Processing of moving images in the geniculocortical pathway. In J. D. Pettigrew, K. J. Sanderson, & W. R. Levick This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. TWO-PROCESS DISTINCTION IN APPARENT MOTION (Eds.), Visual neuroscicnce (pp. 121-144). New York: Cambridge University Press. Pantle, A., Eggleston, R., & Turano, K. (1985). Rules for resolving the ambiguous motion of luminance gratings and amplitude-modulated gratings undergoing abrupt phase shifts. Investigative Ophthalmology & Visual Science, 26(Suppl. 3), 55. Pantle, A., & McCarthy, J. (1988). Two-stage visual motion phenomena. Investigative Ophthalmology & Visual Science, 29(Suppl. 3), 252. Pantle, A. J., & Petersik, J. T. (1980). Effects of spatial parameters on the perceptual organization of a bistable motion display. Perception & Psychophysics, 27, 307-312. Pantle, A. J., & Picciano, L. (1976). A multistable movement display: Evidence for two separate motion systems in human vision. Science, 193. 500-502. Pantle, A., & Turano, K.. (1986). Direct comparisons of apparent motions produced with luminance, contrast-modulated (CM), and texture gratings. Investigative Ophthalmology & Visual Science, MSuppl. 3), 141. Petersik, J. T. (1984). The perceptual fate of letters in two kinds of apparent movement displays. Perception & Psychophysics, 36, 146-150. Petersik, J. T. (1987). Dependence of apparent movement of a subjective figure on the perceptual fate of inducing elements. Perception, 16, 453-459. Petersik, J. T. (1989). Global cooperativity of the short-range process produced with contour-containing stimuli. Manuscript submitted for publication. Petersik, J. T, & Grassmuck, J. (1981). High fundamental spatial frequencies and edges have different perceptual consequences in the 'group/end-to-end' movement phenomenon. Perception, 10, 375382. Petersik, J. T, Hicks, K. 1., & Pantle, A. J. (1978). Apparent movement of successively generated subjective figures. Perception. 7, 371-383. Petersik, J. T., & McDill, M. (1981). A new bistable motion illusion based upon 'kinetic optical occlusion.' Perception, 10, 563-572. Petersik, J. T, & Pantle, A. J. (1979). Factors controlling the competing sensations produced by a bistable stroboscopic motion display. Vision Research, 19, 143-154. Petersik, J. T, Pufahl, R., & Krasnoff, E. (1983). Failure to find an absolute retinal limit of a putative short-range process in apparent motion. Vision Research, 23, 1663-1670. Post, R. B., & Leibowitz, H. W. (1985). A revised analysis of the role of efference in motion perception. Perception, 14, 631-643. Prazdny, K. (1986a). Three-dimensional structure from long-range apparent motion. Perception, 15, 619-625. Prazdny, K. (1986b). What variables control (long-range) apparent motion? Perception, 15. 37-40. Ramachandran, V. S., & Anstis, S. M. (1983). Displacement thresholds for coherent apparent motion in random dot-patterns. Vision Research, 23, 1719-1724. Ramachandran, V. S., & Anstis, S. M. (1986). Figure-ground segregation modulates apparent motion. Vision Research, 26, 1969-1976. 127 Ramachandran, V. S., & Cavanagh, P. (1987). Motion capture anisotropy. Vision Research, 27,97-106. Ramachandran, V. S., Ginsburg, A. P., & Anstis, S. M. (1983). Low spatial frequencies dominate apparent motion. Perception. 12, 457462. Ramachandran, V. S., &Gregory, R. L. (1978). Does colour provide an input to human motion perception? Nature, 275, 55-56. Ramachandran, V. S., Inada, V., & Kiama, G. (1986). Perception of illusory occlusion in apparent motion. Vision Research. 26, 17411749. Ritter, A. B., & Breitmeyer, B. (1987, November). Effects ofdichoptic viewing on bistable motion percepts. Paper presented at the 28th Annual Meeting of the Psychonomic Society, Seattle, WA. Rock, I. (1983). The logic of perception. Cambridge, MA: MIT Press. Sekuler, R., Pantle, A., & Levinson, E. (1978). Physiological basis of motion perception. In R. Held, H. W. Leibowitz, & H.-L. Teuber (Eds.), Handbook of sensory physiology: Vol. 8. Perception (pp. 6796). New York: Springer-Verlag. Shaw, G., &. Ramachandran, V. S. (1982). Interpolation during apparent motion. Perception, 11, 491-494. Swanston, M. T, Wade. N. J., & Day, R. H. (1987). The representation of uniform motion in vision. Perception, 16, 143-159. Ternus, J. (1938). The problem of phenomenal identity. In W. D. Ellis (Ed.), A source book of (Jestall psychology. London: Routledge & Kegan Paul. Ullman, S. (1978). Two dimensionality of the correspondence process in apparent motion. Perception, 7, 683-693. Ullman, S. (1979). The interpretation of visual motion. Cambridge, MA: MIT Press. Ullman, S. (1980). The effect of similarity between line segments on the correspondence strength in apparent motion. Perception, 9, 617-626. van Doom, A. J., Koenderink, J. J., & van de Grind, W. A. (1985). Perception of movement and correlation in stroboscopically presented noise patterns. Perception, 14, 209-224. von Grunau, M. W. (1978a). Dissociation and interaction of form and motion information in the human visual system. Vision Research, IS, 1485-1489. von Grunau, M. W. (1978b). Interaction between sustained and transient channels: Form inhibits motion in the human visual system. Vision Research. /8, 197-201. von Grunau, M. W. (1979). Form information is necessary for the perception of motion. Vision Research. 19. 839-841. von Grunau, M. W. (1986). A motion aftereffect for long-range stroboscopic apparent motion. Perception & Psychophysics, 40, 31.-38. Watson, A. B. (1986). Apparent motion occurs only between similar spatial frequencies. Vision Research, 26, 1727-1730. Watson, A. B., & Ahumada, A. J. (1985). Model of human visual-motion sensing. Journal of the Optical Society of America, A2, 322-342. Williams, D. W, & Sekuler, R. (1984). Coherent global motion percepts from stochastic local motions. Vision Research, 24, 55-62. Received December 21, 1987 Revision received July 8, 1988 Accepted October 5, 1988 •