Image Enhancement 101

import cv2
import matplotlib.pyplot as plt
import numpy as np

# Opening the image.

a = cv2.imread('../Figures/angiogram1.png')
# gamma is initialized.
gamma = 0.5
# b is converted to type float.
b1 = a.astype(float)
# Maximum value in b1 is determined.
b3 = np.max(b1)
# b1 is normalized
b2 = b1/b3
# gamma-correction exponent is computed.
b4 = np.log(b2)*gamma
# gamma-correction is performed.
c = np.exp(b4)*255.0
# c is converted to type int.
c1 = c.astype(int)
# Displaying c1
plt.imshow(c1)

Figure 5.5(a) is an image of the angiogram of blood vessels. The

image is too bright and it is quite difficult to distinguish the blood
vessels from background. Figure 5.5(b) is the image after gamma cor-
rection with γ = 0.5; the image is brighter compared to the original
image. Figure 5.5(c) is the image after gamma correction with γ = 5;
this image is darker and the blood vessels are visible.
102 Image Processing and Acquisition using Python

(a) Input image. (b) Gamma corrected image with

γ = 0.5.

(c) Gamma-corrected image with

γ = 5.

FIGURE 5.5: An example of power law transformation.

5.5 Log Transformation

Log transformation is used to enhance pixel intensities that are
otherwise missed due to a wide range of intensity values or lost at the
expense of high-intensity values. If the intensities in the image range
from [0, L − 1] then the log transformation at (i, j) is given by

t(i, j) = k log(1 + I(i, j)) (5.7)

 Image Enhancement 103
L−1
where k = and Imax is the maximum magnitude value
log(1 + |Imax |)
and I(i, j) is the intensity value of the pixel in the input image at
(i, j). If both I(i, j) and Imax are equal to L − 1, then t(i, j) = L − 1.
When I(i, j) = 0, since log(1) = 0 will give t(i, j) = 0. While the end
points of the range get mapped to themselves, other input values will
be transformed by the above equation. The log can be of any base;
however, the common log (log base 10) or natural log (log base e)
are widely used. The inverse of the above log transformation when the
x
base is e is given by t−1 (x) = e k − 1, which does the opposite of the
log transformation.
Similar to the power law transformation with γ < 1, the log trans-
formation also maps a small range of dark or low-intensity pixel val-
ues in the input image to a wide range of intensities in the output
image, while a wide range of bright or high-intensity pixel values in
the input image get mapped to narrow range of high intensities in the
output image. Considering the intensity range is between [0, 1], Figure
5.6 illustrates the log and inverse log transformations.
The Python code for log transformation is given below.

import cv2
import numpy, math

# Opening the image.

a = cv2.imread('../Figures/bse.png')
# a is converted to type float.
b1 = a.astype(float)
# Maximum value in b1 is determined.
b2 = numpy.max(b1)
# Performing the log transformation.
c = (255.0*numpy.log(1+b1))/numpy.log(1+b2)
# c is converted to type int.
c1 = c.astype(int)
# Saving c1 as logtransform_output.png.
cv2.imwrite('../Figures/logtransform_output.png', c1)
104 Image Processing and Acquisition using Python

FIGURE 5.6: Graph of log and inverse log transformations.

Figure 5.7(a) is a backscattered electron microscope image. Notice

that the image is very dark and the details are not clearly visible.
Log transformation is performed to improve the contrast, to obtain the
output image shown in Figure 5.7(b).

5.6 Histogram Equalization

The histogram of an image was discussed in Chapter 3, “Image and
its Properties.” The histogram of an image is a discrete function, its
input is the gray-level value and the output is the number of pixels with
that gray-level value and can be given as h(xn ) = yn . In a grayscale
image, the intensities of the image take values between [0, L − 1]. As
discussed earlier, low gray-level values in the image (the left side of the
 Image Enhancement 105

(a) Input (b) Output

FIGURE 5.7: Example of log transformation. Original image reprinted

with permission from Mr. Karthik Bharathwaj.

histogram) correspond to dark regions and high gray-level values in the

image (the right side of the histogram) correspond to bright regions.
In a low-contrast image, the histogram is narrow, whereas in an
image with better contrast, the histogram is spread out. In histogram
equalization, the goal is to improve the contrast of an image by rescaling
the histogram so that the histogram of the new image is spread out
and the pixel intensities range over all possible gray-level values. The
rescaling of the histogram will be performed by using a transformation.
To ensure that for every gray-level value in the input image there is a
corresponding output, a one-to-one transformation is required; that is,
every input has a unique output. This means the transformation should
be a monotonic function. This will ensure that the transformation is
invertible.
Before histogram equalization transformation is defined, the follow-
ing should be computed:

• The histogram of the input image is normalized so that the range

of the normalized histogram is [0, 1].

• Since the image is discrete, the probability of a gray-level value,

denoted by px (i), is the ratio of the number of pixels with a gray
106 Image Processing and Acquisition using Python

value i to the total number of pixels in the image. This is generally

called the probability distribution function (PDF).

• The cumulative distribution function (CDF) is defined as C(i) =

Xi
px (j), where 0 ≤ i ≤ L − 1 and where L is the total number
j=0
of gray-level values in the image. The C(i) is the sum of all the
probabilities of the pixel gray-level values from 0 to i. Note that
C is an increasing function.

The histogram equalization transformation can be defined as fol-

lows:
 
C(u) − Cmin
h(u) = round ∗ (L − 1) (5.8)
1 − Cmin
where Cmin is the minimum value in the cumulative distribution. For
a grayscale image with range between [0, 255], if C(u) = Cmin then
h(u) = 0. If C(u) = 1 then h(u) = 255. The integer value for the
output image is obtained by rounding Equation 5.8.
Let us consider an example to illustrate the probability, CDF, and
histogram equalization. Figure 5.8 is an image of size 5 by 5. Let us
assume that the gray levels of the image range from [0, 255].
The probabilities, CDF as C for each gray-level value along with the
output of histogram equalization transformation, are given in Figure
5.9.
The Python code for histogram equalization is given below. The
image is read and a flattened image is calculated. The histogram and
the CDF of the flattened image are then computed. The histogram
equalization is then performed according to Equation 5.8. The flattened
image is then passed through the CDF function and then reshaped to
the original image shape.

import cv2
import numpy as np