Browse Books

Reviewer: Keith E. Price

Textures, in computer vision, are the patterns and variations in an image that are smaller than the objects being analyzed, such as those caused by bricks in a wall, variations within the individual bricks, or grass in a lawn. Rao discusses a taxonomy for texture description and a technique for analysis of oriented textures. Since the book is based on the author's PhD dissertation, it has the advantages and flaws inherent in any dissertation. The majority of the book is concerned with the analysis of oriented textures. These patterns include the grain texture in wood and the flow patterns that arise in analysis of fluids, wind patterns, and so on. The goal of the research is to determine the direction of the oriented texture at different scales. To understand the need for this, consider the patterns in an image of herringbone weave material: at the highest resolution the diagonal patterns of the threads are the strongest, but at lower resolutions the vertical pattern caused by the light and dark areas of the cloth is most apparent. Thus the dominant orientation is different at different scales. From the flow directions, a high-level symbolic description of the texture or flow pattern is generated. This description is in terms of a few descriptors (saddle, spiral, and node) that describe the appearance of the pattern. The remainder of the book discusses the taxonomy of textures; much of the discussion is a review of past work in the field. First, strongly ordered textures are characterized by basic primitives arranged according to some placement rules (such as tiles on a wall or floor). Next, Rao discusses disordered textures, especially in terms of the fractal dimension. The final set of textures are compositional; these are created by combining the earlier primitive textures (strongly ordered, disordered, and weakly ordered or oriented). Using this final taxonomy, the author classifies all the textures from Brodatz's standard collection of texture images [1]. As this is a dissertation, many issues of texture analysis are ignored and others are covered in great depth. The research on the analysis of oriented patterns is important, and the taxonomy provides a means for describing texture patterns. The references to other work in the field are good, as are the discussions of the contributions of these other researchers. Unlike some author-formatted books, this volume has few typographic problems. Figure 5.2(c) is a uniform gray image rather than an image of birch wood, the second paragraph in Section 1.1.1 has several serious problems with punctuation, and the placement of quotation marks relative to commas and periods is inconsistent and usually incorrect (this error is typical of computer programmers who expect quotation marks to have a very different meaning than they do in English).