Fall, 2022

Digital Image Processing

NaïveImage
BayesEnhancement
Analysis(單純貝氏)
(I)
Dr. Tun-Wen Pai
1) Concept of image enhancement
2) Intensity Transformation
3) Histogram Transformation
4) Histogram Equalization
5) Histogram Matching (Specification)

Sept. 19, 2022

 Image Enhancement
• To process an image so that output is “visually better”
than the input, for a specific application.
• Enhancement is therefore, very much dependent on
the particular problem/image at hand.
• Enhancement can be done in either:
□ Spatial domain: operate on the original image
g(x,y) = T[f(x,y)]
□ Frequency domain: operate on the DFT of the original
image
G(u, v) = T[F(u, v)], where
F(u, v) = F [f(x,y)], and G(u, v) = F [g(x,y)]
Image Enhancement Techniques

Point Mask Transform Coloring

Operations Operations Operations Operations

•Image Negative •Smoothing •Low pass •False coloring

•Contrast operations filtering •Full color
Stretching •Median •High pass processing
•Compression of Filtering filtering
dynamic range •Sharpening •Band pass
•Graylevel operations filtering
slicing •Derivative •Homomorphic
•Image operations filtering
Subtraction •Histogram •Histogram
•Image operations operations
Averaging
•Histogram
operations
 Point Operation
• Output pixel value g(x,y) at pixel (x,y) depends only on the input
pixel value at f(x,y) at (x,y) (and not on the neighboring pixel
values).
• We normally write s = T(r), where s is the output pixel value and r is
the input pixel value.

• T is any increasing function that maps [0, 1] into [0, 1].

 Chapter 3
Image Enhancement in the
Spatial Domain

FIGURE 3.4 (a)

Original digital
mammogram.
(b) Negative image
obtained using the
negative
transformation in
Eq. (3.2-1).
(Courtesy of G.E.
Medical Systems.)
 Comprssion of Dynamic Range
• When the dynamic range of the input grayvlauels is large compared
to that of the display, we need to “compress” the grayvalue range ---
example: Fourier transform magnitude.
• Typically we use a log scale.

s  T (r )  c log(1 | r |)
Gamma correction:
s1  0, s 2  1, and
 0, r  r1 
 g 
 r  r1  
T(r)    , r1  r  r2 
 r2  r1  
 1, r  r 
 2 
Original image Gamma Correction Original image as viewed on a monitor with
a gamma of 2.5

Gamma-corrected image Gamma-corrected image as viewed on the

same monitor
 Contrast Stretching
• Increase the dynamic range of grayvalues in the input image.
• Suppose you are interested in stretching the input intensity values in the
interval [r1 , r2]:

• Note that (r2-r1) < (s2 –s1). The grayvalues in the range [r1 , r2] is stretched
into the range [s1, s2]
• Special cases:
□ Thresholding or binarization
r1 = r2, , s1 = 0 and s2 = 1
 Image Enhancement: Histogram-
based methods
• The histogram of a digital image with grayvalues r0 ,r1 ,…, rL-1 is the
discrete funciotn
# pixels with value rk
nk
p(rk ) 
n Total # pixels in image
• The function p(rk) represents the fraction of the total number of pixels with
grayvalue rk.

• Histogram provides a global description of the appearance of the image.

• If we consider the grayvalues in the image as realizations of a random

variable R, with some probability density, histogram provides an
approximation to this probability density. In other words,
p r [ R  rk ]  p(rk )
 Some typical Histograms
• The shape of a histogram provides useful information for contrast
enhancement.
 Histogram Equalization
• Let us assume for the moment that the input image to be enhanced
has continuous grayvalues, with r = 0 representing black and r = 1
representing white.
• We need to design a grayvalue transformation s = T(r), based on the
histogram of the input image, which will enhance the image.
• As before, we assume that:
□ T(r) is a monotonically increasing function for 0  r  1
(preserves order from black to white)
□ T(r) maps [0, 1] into [0, 1] (preserves the range of allowed
grayvalues).
FIGURE 3.16
A gray-level
transformation
function that is both
single valued and
monotonically
increasing.
 Histogram Equalization
• Let us denote the inverse transformation by r = T-1(s). We assume that the
inverse transformation also satisfies the above two conditions.
• We consider the grayvalues in the input image and output image as random
variables in the interval [0, 1].
• Let pin(r) and pout(s) denote the probability density of the grayvalues in the
input and output images.
• If pin(r) and T(r) are known, and T-1(s) satisfies condition 1, we write
(result from probability theory):
dr
pout ( s)  [ pin (r ) ]r T 1 ( s )
ds
• One way to enhance the image is to design a transformation T(.) such that
the grayvalues in the output is uniformly distributed in [0, 1], i.e.

pout ( s)  1, 0  s  1

• In terms of histograms, the ouput image will have al grayvalues in “equal

proportion.”
 Histogram Equalization
• Consider the transformation
r
s  T (r )   pin ( w)dw, 0  r  1
0

• Note that this is the cumulative distribution function (CDF) if pin(r) and
satisfies the previous two conditions.

• From the previous equation and using the fundamental theorem of

calculus,
ds
 pin (r )
dr
• Therefore, the output histogram is given by
1
pout ( s)  [ pin (r ) ] 1
pin (r ) r T ( s )
 [1]r T 1 ( s )  1, for 0  s  1

• The output probability density function is uniform, regardless of the input.

An example of histogram equalization in a continuous pdf.
An example of histogram equalization in a discrete pdf.
 Histogram Equalization
• Thus, using a transformation function equal to the CDF of input grayvalues
r, we can obtain an image with uniform grayvalues.

• This usually results in an enhanced image, with an increase in the dynamic

range of pixel values.

• For images with discrete grayvalues, we have

nk
pin (rk )  , for 0  rk  1, and 0  k  L  1
n
□ L: Total number of graylevels
□ nk:Number of pixels with grayvalue rk
□ n: Total number of pixels in the image

• The discrete version of the previous transformation based on CDF is given

by: k n k
sk  T (rk )     pin (rj ), for 0  k  L - 1
j

j 0 n j 0
 Histogram Specification
• Histogram equalization yields an image whose pixels are (in theory) uniformly
distributed among all graylevels.

• Sometimes, this may not be desirable. Instead, we may want a transformation that
yields an output image with a pre-specified histogram. This technique is called
histogram specification.

• Again, we will assume, for the moment, continuous-grayvalues.

• Suppose, the input image has probability density pin(r). We want to find a
transformation z = H(r), such that the probability density of the new image obtained
by this transformation is pout(z), which is not necessarily uniform.

• First apply the transformation

r
s  T (r )   pin ( w)dw, 0  r  1 (*)
0
This gives an image with a uniform probability density.

• If the desired output image were available, then the following transformation would
generate an image with uniform density
 z
v  G( z )   pout ( w)dw, 0  z  1 (**)
0

• From the grayvalues v we can obtain the grayvalues z by using the inverse
tranformation , z = G-1(v).

• If instead of using the grayvalues v obtained from (**), we use the grayvalues s
obtained from (*) above (both are uniformly distributed!), then the point
transformation will generate an image with the specified density p out(z), from an
input image with density pin(r)!

• For discrete graylevels, we have

k nj
sk  T (rk )   , for 0  k  L - 1 and
j 0 n
k
vk  G ( zk )   pout ( z j ), for 0  k  L - 1
j 0

• zk = G-1 (vk) = G-1(sk) = G-1( T(rk) )

• Dealing with a small set of discrete grayvalues.

(a) Image of the Mars moon Photos taken by NASA’s Mars Global
Surveyor. (b) Histogram. (Original image courtesy of NASA.)
(a) Transformation
function for
histogram
equalization.
(b) Histogram-
equalized image
(note the washed-out
appearance).
(c) Histogram of (b).
(a) Specified histogram.
(b) Curve (1) is from Eq.(3.3-
14), using the histogram in (a);
curve (2) was obtained using the
iterative procedure in Eq.(3.3-
17).
(c) Enhanced image using
mappings from curve (2).
(d) Histogram of (c).
SEM image of a tungsten
filament and support, magnified
approximately 130X. (Original
image courtesy of Mr. Michael
Shaffer, Department of
Geological Sciences, University
of Oregon, Eugene).
(a) Image formed from all local means obtained from Fig. 3.24 using Eq.(3.3-21). (b)
Image formed from all local standard deviations obtained from Fig. 3.24 using
Eq.(3.3-22). (c) Image formed from all multiplication constats used to produce the
enhanced image shown in Fig. 3.26.
Enhanced SEM
image. Compare with
Fig.3.24. Note in
particular the
enhanced area on the
right side of the
image.
 Enhancement by
image subtraction.
(a) Mask image.
(b) An image
(taken after
injection of a
contrast medium
into the
bloodstream) with
mask subtracted
out.

Image Subtraction
• In this case, the difference between two “similar” images is compute to
highlight or enhance the differences between them:

g (m, n)  f1 (m, n)  f 2 (m, n)

• It has applications in image segmentation and enhancement
Image Averaging for noise reduction

• Noise is any random (unpredictable) phenomenon that contaminates an

image.

• Noise is inherent in most practical systems:

□ Image acquisition
□ Image transmission
□ Image recording

• Noise is typically modeled as an additive process:

g (m, n)  f (m, n)  (m, n)

Noisy Image Noise-free Noise

Image
• The noise η(m, n) at each pixel (m, n) is modeled as a random variable.

• Usually, η(m, n) has zero-mean and the noise values at different pixels are
uncorrelated.
• Suppose we have M observations {gi (m, n)}, i=1, 2, …, M, we can (partially)
mitigate the effect of noise by “averaging”
M
1
g (m, n) 
M
 g (m, n)
i 1
i

• In this case, we can show that

E[ g (m, n)]  f (m, n)

1
Var[ g (m, n)]  Var[ (m, n)]
M
• Therefore, as the number of observations increases (M → ∞), the effect of
noise tends to zero.
(a) Image of
Galaxy Pair
NGC3314. (b)
Image corrupted by
additive Gaussian
noise with zero
mean and a
standard deviation
of 64 gray levels.
(c)-(f) Results of
averaging K = 8,
16, 64, and 128
noisy images.
(Original image
courtesy of
NASA).
(a) From top to
bottom:
Difference images
between Fig.3.30
(a) and the four
images in
Figs.3.30(c)
through (f),
respectively. (b)
Corresponding
histograms.