0% found this document useful (0 votes)
53 views6 pages

Optik: Bhupendra Gupta, Mayank Tiwari

Uploaded by

mayank tiwari
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
53 views6 pages

Optik: Bhupendra Gupta, Mayank Tiwari

Uploaded by

mayank tiwari
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 6

Optik 127 (2016) 1671–1676

Contents lists available at ScienceDirect

Optik
journal homepage: www.elsevier.de/ijleo

Minimum mean brightness error contrast enhancement of color


images using adaptive gamma correction with color preserving
framework
Bhupendra Gupta ∗ , Mayank Tiwari
Indian Institute of Information Technology, Design & Manufacturing Jabalpur, 482005 MP, India

a r t i c l e i n f o a b s t r a c t

Article history: In this paper, we propose an efficient method to enhance contrast of digital images. Image contrast
Received 25 December 2014 enhancement is a pre-processing step that improves efficiency of other image dealing applications such
Accepted 11 October 2015 as computer vision and pattern recognition. The proposed method enhances contrast and brightness
of given image using the gamma correction and weighted probability distribution of luminance pixels.
Keywords: Experiments results show that the proposed method is able to enhance contrast of all type of color images
Contrast enhancement
without much affecting its visual and color information.
Gamma correction
© 2015 Elsevier GmbH. All rights reserved.
Histogram equalization
Histogram modification
Color preserving framework

1. Introduction of implementation and computationally less expensiveness than


other methods.
Insufficient contrast in back-light images reduce the ability of Although the HE based methods are widely used, in his-
observer in analyzing the image. In many cases insufficient contrast togram equalization we do not have any mechanism to control
may produce difficulties in subsequent processing. This insufficient the enhancement level, due to this sometimes output image may
contrast may occur due to various reasons such as presence of have over enhanced regions. Also HE could not effectively work,
clouds in the captured scene, lack of operator expertise and inade- when the input image contains regions that are significantly darker
quacy of the image capture device, etc. In the available literature we or brighter than other parts of the image. Another problem with
are giving introduction of various methods that are used for image HE is, in theory it is clearly shown that mean brightness of his-
contrast enhancement. These methods can be broadly divided into togram equalize image is always middle gray level; regardless the
two categories they are, direct enhancement [2–4] and indirect mean brightness of input image. This problem is known as the
enhancement [5–7]. In direct enhancement methods, the image ‘mean-shift’ problem. In [8], K. Zuiderveld proposed a widely used
contrast can be directly defined by a specific contrast term [2–4]. method known as ‘contrast limited adaptive histogram equaliza-
However, most of these metrics cannot simultaneously gauge the tion’ (CLAHE). The CLAHE method was able to reduce the over
contrast of simple and complex patterns in images which contain enhancement of contrast in the processed image but it was not
both [4]. capable in reducing the mean brightness change in the processed
Histogram equalization (HE) based techniques fall under the image. To solve the ‘mean-shift’ problem, Yeong-Taeg in [9] pro-
category of indirect enhancement methods. These techniques posed ‘brightness preserving bi-histogram equalization’ (BBHE)
attempt to enhance image contrast by redistributing the proba- where the image X is divided into two sub-images XL and XU
bility density [1]. Using these techniques [9–15] the intensities based on mean XM of brightness of input image and then equal-
can be better redistributed within the dynamic range. Histogram ize these two sub-histograms separately. Later Zhang in [10]
equalization techniques [9–15] are popular due to their simplicity proposed ‘equal area dualistic sub-image histogram equalization’
(DSIHE), though this approach also separates an input image
into two sub-sections, the only difference between BBHE and
∗ Corresponding author. Tel.: +91 9425155354.
DSIHE is, in later method the separation is based on the median
E-mail addresses: [email protected] (B. Gupta),
value.
[email protected] (M. Tiwari). An extension of BBHE has been proposed by Chen and Ramli
URL: http://https://www.bhupendragupta.com (B. Gupta). in [11], known as ‘minimum mean brightness error bi-histogram

http://dx.doi.org/10.1016/j.ijleo.2015.10.068
0030-4026/© 2015 Elsevier GmbH. All rights reserved.
1672 B. Gupta, M. Tiwari / Optik 127 (2016) 1671–1676

equalization’ (MMBEBHE). This method provides maximum bright- 3. The proposed method
ness preservation. The MMBEBHE proposes to perform the
separation based on the threshold level, which would yield min- The resultant cdf curve of AGCWD [16] method is shown in Fig. 1.
imum absolute mean brightness error [11]. The MMBEBHE is a It is clear from Fig. 1 that for ˛ = 1.2 and ˛ = 0.8 the resultant
useful tool to control the brightness difference between input transformation curve is not increasing smoothly. To prove our this
and processed image. In [12], Chen and Ramli proposed another claim (in image1 for transformation curve ˛ = 1.2) we have marked
interesting method called ‘recursive mean-separate histogram two intervals on x-axis (x1 , x2 ); (x3 , x4 ) and their corresponding
equalization’ (RMSHE), here authors suggested recursive division intervals on y-axis (y1 , y2 ); (y3 , y4 ). The small interval (x1 , x2 ) is
of histograms, based on the local mean. In each recursive step transformed to sufficiently large interval (y1 , y2 ) and in this case
existing sub-histogram is divided into two sub-histograms. After sufficient weights will be assigned to Eq. (3). However on the other
rth recursion the number of sub-histogram is 2r , where number of hand due to non-smooth nature of the curve the large interval (x3 ,
recursion depends on choice of user. Also authors proved math- x4 ) is transformed to a small interval (y3 , y4 ) and hence weights for
ematically that as r increases, the mean brightness of processed a large interval will be not sufficiently large. This may also lead to
image approaches towards the mean brightness of input image. Sim loss of information in processed image.
et al. in [13], improved DSIHE into ‘recursive sub-image histogram To solve this problem and to make the resultant transformation
equalization’ based contrast enhancement (RSIHE), by introduc- curve to be smoothly increasing with sufficient color preservation;
ing recursive segmentation in the similar manner as Chen and we proposed a new method called ‘adaptive gamma correction with
Ramil proposed in [12], this method is similar to RMSHE but it color preserving framework’ (AGCCPF). The proposed method uses
uses median values instead of mean values to divide histogram into two-steps processing; initially it enhances contrast and brightness
sub-histograms. of given image using modified probability distribution of luminance
Recently, S.C. Huang et al. [16], propose ‘efficient contrast pixels followed by the gamma correction. In second step it uses a
enhancement using adaptive gamma correction with weighting color preserving framework for color restoration.
distribution’ (AGCWD). This method is an automatic transforma- The objective of all histogram based contrast enhancement
tion technique that improves the brightness of dimmed images methods is to transform the histogram of an input image closer
via the gamma correction and probability distribution of lumi-
nance pixels; also this method produces enhanced images of
comparable or higher quality than those produced by other
methods [16].
The organization of this work is as follows. After introduc-
ing about research background in Section 1, Section 2 describes
the AGCWD method. Section 3 explains the proposed method in
details. For evaluation of proposed method experimental results are
shown in Section 4. Finally Section 5 summarizes and concludes
the proposed work.

2. The AGCWD method

As mentioned earlier in [16], S.C. Huang et al. proposes ‘effi-


cient contrast enhancement using adaptive gamma correction with
weighting distribution’, (AGCWD). The AGCWD method can pro-
gressively increase the low intensity and avoid the significant
decrement of the high intensity. The AGCWD method applies the
weighting distribution (WD) function to slightly modify the statis-
tical histogram and lessen the generation of adverse effects. The
WD function is formulated as:
 pdf (l) − pdf ˛
min
pdf w (l) = pdf max , (1)
pdf max − pdf min

where ˛ is the adjusted parameter, pdfmax is the maximum pdf of


the statistical histogram, and pdfmin is the minimum pdf. Based on
Eq. (1) the gamma parameter is given as:

 = 1 − cdf (l), (2)

where cdf is resultant cumulative distribution function calculated


after normalizing pdf w . The adaptive gamma correction (AGC) used
by S.C. Huang et al. in [16] is given as:

 l   l 1−cdf (l)
T (l) = (L − 1) = (L − 1) , (3)
L−1 L−1

where (L − 1) is the maximum intensity value of the


image. It is clear that slight increase in cdf(l) will cause Fig. 1. Resultant transformation curve of AGCWD method, after applying weighting
distribution function on image1 , image2 and image3 . Here x-axis shows gray level of
drastic change in intensity value l of the processed
image and y-axis shows corresponding cumulative distribution function at each gray
image. level.
B. Gupta, M. Tiwari / Optik 127 (2016) 1671–1676 1673

to uniform histogram, and also the objective is to maximize the


image contrast by transforming the gray level distribution of input
image to a gray level distribution which is closer to uniform gray
level distribution.
Here, we use a histogram modification function for contrast
enhancement. This function has the ability to control the level of
contrast enhancement in the processed image.
In histogram equalization we consider an image as a 2 dimen-
sional array of gray levels. Let the (i, j) element of this array is X(i;j)
be the intensity of (i, j) pixel of the image, where X(i;j) is from the
L discrete gray levels denoted by {X0 , X1 , . . ., XL−1 }. Let H be the
histogram of the input image and let Hu be the uniform histogram.
The main idea is to obtain a modified histogram H́ from the input
histogram H such that the difference between H́ and Hu is suffi-
ciently small, keeping the H́ closer to H. Now it is an optimization
problem and can be formalized as:

H́ = H + (1 − )Hu , (4)

where 0 ≤  ≤ 1.
In Fig. 2, we are showing mapping function for different values of
. This mapping function is nothing but the cumulative distribution
function of resultant modified histogram. For  = 0.5 the mapping
curve increases smoothly to the maximum value and this causes
no over enhancement in the processed image. Now based on Eq.
(4) the corresponding normalized probability mass function (PMF)
can be given as:

H́(l)
pmf (l) = , (5)
H́
L−1
where H́ = k=0
H́(k). Now based on Eq. (5) the corresponding
cdf is given as:


l
cdf (l) = pmf (k). (6)
k=0

The proposed method directly utilizes the cdf (which is generated


after histogram modification function). This cdf is applied in a nor- Fig. 2. Resultant transformation curve after applying histogram modification func-
tion. Here the transformation curves are shown for image1 , image2 and image3 .
malized gamma function which modifies the transformation curve.
Here x-axis shows gray level of image and y-axis shows corresponding cumulative
After calculation of cdf using Eq. (6) we apply adaptive gamma distribution function at each gray level.
correction using Eqs. (2) and (3).
By the histogram modification function we can adjust the cdf
of input image so that the resultant cdf is smoothly increasing and the color information of input image and applying it in processed
hence no over enhancement will occur. However when we deal image. Now we can restore color information in processed image
with color images then our aim is not only to enhance their contrast by the following:
but also to preserve their color information up to some extent. For
preservation of color information in processed image, we are using PRC IMG = ı(INT IMG ) + (1 − ı)(INP IMG ), (7)

Fig. 3. Set of images used in the proposed work, images are given names as image1 , image2 , up to image12 .
1674 B. Gupta, M. Tiwari / Optik 127 (2016) 1671–1676

where INTIMG is image obtained after applying Eq. (3), INPIMG is illumination due to back-light effect. Generally change in illumi-
input image and PRCIMG is corresponding processed image. Param- nation can be due to many factors, such as location of light source,
eter 0 ≤ ı ≤ 1. It is clear that for ı = 0 PRCIMG = INPIMG and in this case cloud cover, intensity of sunshine, etc. Fig. 3 shows set of images,
there would be maximum color preservation without any contrast used by us in the proposed work.
enhancement in the PRCIMG . For ı = 1 PRCIMG = INTIMG and in this Fig. 4 shows enhancement results of various methods for image2 .
case there would be minimum color preservation with contrast The image2 contains scene with an object and clear sky background.
enhancement. The HE method produces over enhancement and noise enhance-
In the proposed work we are working with color images using ment too; in the processed image (b). Results of other methods
the RGB color model. The RGB color model is an additive color model such as RMSHE and RSIHE are able to enhance the contrast of the
in which red, green, and blue light are added together in various image at the cost of natural appearance’s loss. The CLAHE method
ways to reproduce a broad array of colors. The name of the model is producing better results but this method is not able to maintain
comes from the initials of the three additive primary colors, red, the mean brightness in the processed image. The AGCWD method
green, and blue. In implementation part of proposed algorithm, we produces good results but in this case color-information of the sky
are applying the algorithm on each color channel of given color in the background is washed out. On the other hand the AGCCPF
image, after that we are combining all three channels to form the method is able to enhance contrast of given image without much
processed image. affecting it color-information and natural appearance.
Table 1 shows results of Absolute Mean Brightness Error (AMBE).
4. Experimental results AMBE is used to calculate difference in mean brightness between
two images. Mathematical expression to calculate AMBE between
In this section, we demonstrate performance of the proposed two images is given as:
method in comparison with some existing HE based contrast
AMBE = |XM − YM |, (8)
enhancement methods, like HE, RMSHE, RSIHE, RSWHE and
AGCWD. For RMSHE, RSIHE and RSWHE we are using recursion where XM and YM are mean brightness of input and processed image
level r = 1. As test images we are taking images having poor respectively.
Based on results of Table 1, we observe that the proposed
method has least values in all 12 images as compare to other meth-
ods. Further if we look at last row of Table 1, which shows average

Fig. 4. Enhancement results of various methods for image2 . Here (a) input image, Fig. 5. Enhancement results of proposed method for image1 , image1 2, image5 and
(b) result of HE, (c) result of RMSHE (r = 1), (d) result of RSIHE (r = 1), (e) result of image9 . Here first column shows the input images second column shows corre-
CLAHE (AMBE = 6.514), (f) result of AGCWD (˛ = 0.8), (g) proposed method ( = 0.4 sponding processed images. For these 4 images we have considered  = 0.4 and
and ı = 0.4) (AMBE = 0.011), (h) proposed method ( = 0.25 and ı = 0.4). ı = 0.4.
B. Gupta, M. Tiwari / Optik 127 (2016) 1671–1676 1675

Table 1
Absolute Mean Brightness Error (AMBE).

Image name HE RMSHE RSIHE CLAHE AGCWD AGCCPF

image1 56.834 8.750 4.380 17.055 33.376 0.024


image2 35.338 1.784 8.324 6.514 29.170 0.011
image3 58.271 7.797 7.088 5.656 22.077 0.036
image4 40.030 5.952 1.411 13.893 32.549 0.070
image5 63.714 4.504 14.058 21.957 40.720 0.001
image6 2.382 5.522 3.010 7.077 33.263 0.026
image7 42.255 7.026 1.705 20.659 36.037 0.024
image8 119.163 24.613 50.735 22.932 48.232 0.029
image9 3.278 10.831 18.494 4.334 27.806 0.006
image10 50.577 8.043 3.217 17.511 34.718 0.001
image11 37.607 3.070 4.218 16.417 31.385 0.005
image12 13.849 10.385 16.100 0.298 30.044 0.018

Average 43.608 8.190 11.062 12.858 33.281 0.021

results of AMBE then we find that the proposed method has least where MSE is Mean Square Error and it is defined as:
average AMBE values among other methods. This shows that our
1 
M N
method is able to maintain mean brightness of input image in the 2
MSE = |X(i, j) − Y (i, j)| , (10)
processed image. MN
i=1 j=1
Table 2 shows results of Peak Signal to Noise Ratio matrix. Peak
Signal to Noise Ratio (PSNR) is widely used matrix which is used to where X(i, j) and Y(i, j) denotes intensity at location (i, j) in the input
measure contrast enhancement [13,14,17,18]. Greater PSNR values and processed image respectively.
indicate better the image quality [14]. To calculate PSNR between Table 2 proves that proposed method enhances more contrast
two images (each image is having L discrete gray levels in the range than HE, RMSHE, RSIHE, CLAHE and AGCWD (as higher PSNR value
{X0 , X1 , . . ., XL−1 }), mathematical expression is given as: indicates better contrast enhancement [13,14,17,18]). On the basis
(L − 1)2
PSNR = 10log10 , (9)
MSE

Fig. 6. Enhancement results of proposed method for image4 , image6 , image7 and Fig. 7. Enhancement results of proposed method for image2 , image1 0, image1 1, and
image8 . Here first column shows the input images second column shows corre- image3 . Here first column shows the input images second column shows corre-
sponding processed images. For these 4 images we have considered  = 0.4 and sponding processed images. For these 4 images we have considered  = 0.4 and
ı = 0.4. ı = 0.4.
1676 B. Gupta, M. Tiwari / Optik 127 (2016) 1671–1676

Table 2
Peak Signal to Noise Ratio (PSNR).

Image name HE RMSHE RSIHE CLAHE AGCWD AGCCPF

image1 11.43 21.25 25.25 19.46 14.92 23.65


image2 13.65 17.67 19.53 21.40 16.53 24.62
image3 10.31 15.80 18.88 22.01 18.90 26.88
image4 13.49 21.68 25.36 19.96 15.21 21.83
image5 9.98 23.19 17.66 18.69 13.18 23.18
image6 23.04 21.68 22.63 17.47 16.29 23.42
image7 13.66 20.33 22.73 17.81 15.20 26.05
image8 5.78 13.95 9.90 18.53 11.45 31.29
image9 18.03 18.28 18.19 21.73 18.26 27.96
image10 12.71 22.28 26.94 19.27 15.46 26.70
image11 13.94 20.31 23.16 18.17 15.78 22.80
image12 19.39 20.15 19.18 16.99 17.37 26.14

Average 13.79 19.71 20.79 19.29 15.71 25.38

of results of Tables 1 and 2 we can conclude that the AGCCPF is a [6] A. Polesel, G. Ramponi, V. Mathews, Image enhancement via adaptive unsharp
good method for contrast enhancement with maximum brightness masking, IEEE Trans. Image Process. 9 (March (3)) (2000) 505–510.
[7] Y.S. Chiu, F.C. Cheng, S.C. Huang, Efficient contrast enhancement using adaptive
preservation and color preservation. gamma correction and cumulative intensity distribution, in: Proc. IEEE Conf.
In the following Figs. 5–7 we are showing results of proposed Syst. Man Cybern., 2011, pp. 2946–2950.
method on all 12 images. [8] K. Zuiderveld, Contrast limited adaptive histogram equalization, in: P.S. Heck-
bert (Ed.), Graphics Gems IV, Academic Press, Cambridge, MA, 1994, pp.
474–485 (chapter VIII).
5. Conclusion [9] Y. Kim, Contrast enhancement using brightness preserving bi-histogram equal-
ization, IEEE Trans. Consum. Electron. 43 (February (1)) (1997) 1–8.
[10] Y. Wan, Q. Chen, B. Zhang, Image enhancement based on equal area dualistic
The AGCCPF method provides optimum contrast enhancement sub-image histogram equalization method, IEEE Trans. Consum. Electron. 45
while preserving the brightness of given color image and hence (February (1)) (1999) 68–75.
this method is suitable for all types of color images. Experimental [11] S. Chen, A.R. Ramli, Minimum mean brightness error bi-histogram equaliza-
tion in contrast enhancement, IEEE Trans. Consum. Electron. 49 (4) (2003)
results show that AMBE of the proposed method is less in compar- 1310–1319.
ison of other methods. Also PSNR of the proposed method is better [12] S. Chen, R. Ramli, Contrast enhancement using recursive mean-separate his-
from HE, RMSHE, RSIHE, CLAHE and AGCWD. On the basis of analy- togram equalization for scalable brightness preservation, IEEE Trans. Consum.
Electron. 49 (4) (2003) 1301–1309.
sis of these two metrics, it is clear that AGCCPF preserves the input
[13] K.S. Sim, C.P. Tso, Y. Tan, Recursive sub-image histogram equalization
image brightness more accurately and gives processed image with applied to gray-scale images, Pattern Recognit. Lett. 28 (July (10)) (2007)
good contrast enhancement. 1209–1221.
[14] M. Kim, M.G. Chung, Recursively separated and weighted histogram equal-
ization for brightness preservation and contrast enhancement, IEEE Trans.
References Consum. Electron. 54 (3) (2008 Aug) 1389–1397.
[15] T. Celik, T. Tjahjadi, Automatic image equalization and contrast enhancement
[1] T. Arici, S. Dikbas, Y. Altunbasak, A histogram modification framework and its using Gaussian mixture modeling, IEEE Trans. Image Process. 21 (January (1))
application for image contrast enhancement, IEEE Trans. Image Process. 18 (2012) 145–156.
(September (9)) (2009) 1921–1935. [16] S.C. Huang, F.C. Cheng, Y.S. Chiu, Efficient contrast enhancement using adaptive
[2] A. Beghdadi, A.L. Negrate, Contrast enhancement technique based on local gamma correction with weighting distribution, IEEE Trans. Image Process. 22
detection of edges, Comput. Vis. Graph. Image Process. 46 (May (2)) (1989) (March (3)) (2013) 1032–1041.
162–174. [17] S. Mohan, M. Ravishankar, Modified contrast limited adaptive histogram equal-
[3] H.D. Cheng, H.J. Xu, A novel fuzzy logic approach to contrast enhancement, ization based on local contrast enhancement for mammogram images, in:
Pattern Recognit. 33 (May (5)) (2000) 809–819. Mobile Communication and Power Engineering Communications in Computer
[4] J. Tang, X. Liu, Q. Sun, A direct image contrast enhancement algorithm in the and Information Science, vol. 296, 2013, pp. 397–403.
wavelet domain for screening mammograms, IEEE J. Sel. Top. Signal Process. 3 [18] S. Mohan, M. Ravishankar, Optimized histogram based contrast limited
(February (1)) (2009) 74–80. enhancement for mammogram images, short paper, ACEEE Int. J. Inf. Technol.
[5] R. Sherrier, G. Johnson, Regionally adaptive histogram equalization of the chest, 3 (March (1)) (2013).
IEEE Trans. Med. Imaging 6 (January (1)) (1987) 1–7.

You might also like