You are on page 1of 4

201O International Conference On Computer Design And Appliations (ICCDA 2010)

An Improved Image Contrast Enhancement in Multiple-Peak Images Based on


Histogram Equalization

Fan Yang, Jin Wu


Information Science and Engineering Department, Wuhan University of Science and Technology (WUST),
Wuhan 430081, China
offeryang@sina.com,wujin1988@163.com

Abstract-The histogram equalization (HE) is one of the most quantized bi-histogram equalization (BHE) [21. Sengee et al
popular methods for image contrast enhancement. However, propose a new method called brightness preserving weight
HE algorithm has two main disadvantages. To solve these clustering histogram equalization (BPWCHE) [31.
problems, this paper presents an improved image contrast Recently, contrast enhancement techniques are widely
enhancement based on histogram equalization, which is used in many fields, there are putting forward new
especially suitable for multiple-peak images. Firstly, the input
requirements and solutions. Mohammad et al raise vessel
image is convolved by a Gaussian filter with optimum
enhancement using directional features [41. Fronthaler et al
parameters. Secondly, the original histogram can be divided
propose local features for enhancement and minutiae
extraction in the fingerprints [51. Gasparini et al offer the low
into different areas by the valley values of the image histogram.

quality images enhancement using visual attention [61. Zhou


Thirdly, using of our proposed method processes images. This
method outperforms others on the aspects of simplicity and
et al propose interactive image enhancement by fuzzy
adaptability. We have applied the proposed algorithm on a I
database which includes 160 normal images and compared our
relaxation [ ll. Wang et al present a real-time contrast
proposed method with HE and RHE. The result demonstrates enhancement algorithm for infrared images based on plateau
1
that the proposed algorithm has good performance in the field histogram [ 21.
of image enhancement. This method can also be realized by Despite the mentioned algorithms can often increase
simple hardware and consumer electronics, efficiently. contrast of the images but they usually bring certain
undesirable effects in the results. In this paper we will
Keywords-Histogram Equalization, Image Processing, Contrast present an efficient algorithm in achieving widely dynamic
Enhancement, Gray Level Transformation range and improving contrast of the images. Firstly we will
apply the Gaussian filter convolved the original images. For
one thing, certain input images are contaminated with
I. INTRODUCTION Gaussian noise in the processing of data collection and
In our daily life, HE techniques are widely used in the transmission, for another the histogram equalization function
filed of consumer electronics, medical image processing, calculates the value that indicates both the lever of pixels and
image matching and searching, speech recognition and all the intensity pixels in the certain neighborhood. Secondly,
texture synthesis because it has high efficiency and the original histogram will be divided into different areas by
simplicity [71 [81 [91 [\01. However, sometimes it also degenerate the valley values of the image histogram. Thirdly, applies
the result which is often called as washed out effect. Figure 1 proposed histogram equalization method on each segment
shows this phenomenon. separately. We have applied the proposed algorithm on a
r- database which includes 160 normal images. This method
outperforms others on the aspects of simplicity and
adaptability.
The organization of this paper is as follows. In section 2,
the proposed algorithm is introduced. In section 3,we present
some of the results that compare the new method with the
traditional histogram equalization. In section 4 presents the
,,' ,,' conclusion.

II. PROPOSED ALGORITHM

A. Gaussian Filter
In order to reduce the noise's interference and improve
the quality of input image, in this work we propose to use
Figure 1 Washed Out Effect Gaussian filter convolving the image firstly. Gaussian filter
In the recently years, many researchers proposed many reduces the difference in brightness between adjacent
useful algorithms to preserve the brightness of images. Jiang elements. It also can reduce blocking effects. As mask size
et al propose a new method, known as local histogram is increased, blocking effects can be decreased. However,
equalization with brightness preservation [l1. Kim proposes computation complexity is increased.

978-1-4244-7164-51$26.00 © 2010 IEEE VI-346 Volume 1


2010 International Conference On Computer Design And Appliations (ICCDA 2010)

Gaussian filter IS calculated by using the following Where N is the total number of pixels in the image, nk is
equation [14]: the number of pixels that have gray level rk, and L is the
total number of possible gray levels in the image.
1 , , (1)
exp(- �)
. + .
G(i,j)= �
...., 21UY 20'
p(rk)
p'(rk) =[ r XPmax k=O ...L-1. (4)
where (j is the standard deviation in the area of mask size. Pmax
(i, j) are the coordinates relative to the center of the filter. Where Pmax is the max value of sub-lays' PDF. If the
The value of (j should be properly chosen. There is
image histogram includes certain high frequency which as
actually a relationship between (j and the size of mask in
[13]
shown in figure2 (b), we will set a suitable value to limit it

R' (2) and we call this suitable value as Pmax . We will use this
0'=
2In(O.OOI) formula to calculate the probability of gray-scale.
where R is the size of Gaussian filter. The default value for The aim is to enlarge the distance between two adjacent
filter size is [3, 3]; the default value for sigma is 0.5. gray levels of the PDF values. Each element of the
proportion in the image can be improved to varying degrees.
B. Image Histogram Segmention Hence it can reduce the difference the shape of probability
To make the processing of the image enhancement more density distribution.
purposed and adaptive, firstly we proposed to analyze the Cumulative Distribution Function (CDF) can also be
image histogram. Image histogram can be divided into computed as equation (5):
several sub-layer images' histogram by local minimum gray i=k
level, as shown in figure2 (a) which is the histogram of a C(rk) = LP(lj) k=O, ,L-1 ... (5)
image. From the image we can be seen that there are three i=O
low points in it, so we can divide the histogram into four We assume C(rk) satisfies the following conditions:
departments. If the histogram as shown in figure2 (b) in (a) C(rk) is single-valued and monotonically increasing in
which there are some gray levels have much more number
than others as plus, we will set a suitable value to the interval
constrained the max frequency of the gray-level. In this (b) O::;C(rk)::;1 for k=O, , L-1
...

way, the proportion of low frequency gray level will be Transformation can be computed as equation (6):
enhanced. This is the preparation for the next step. The
schematic shows in Figure2. T(rk) = ro + (rL-1 - rO)xC(rk) + 0.5 k=O, . , L-1
. . (6)

III. EXPERIMENTS AND RESULTS


In order to evaluate the technology presented in this paper,
we compare the performance of HE, BHE and apply it on a
database which includes 160 normal images that covered all
kinds of images such as medical images, biological images,
Gray-level crowed images and so on.
(a) Two groups' experiments have been carried out in this
Number
paper. The first one investigates the relationship between the
numbers of sub-layer and the performance of our proposed
algorithm.
The first image which is Lena with size of 512x512
pixels is displayed in Fig. 3 (a) and its corresponding
histogram as shown in Fig. 3(b). Fig. 2 (c) presents the
(b) Gray-level
performance of dividing the original histogram into two
Figure 2 Histogram Segmentation Schematic departments. According to the distribution of probability of
gray levels, we set the cut point is 75.Fron the Fig. 3 (d), we
can see the scope of image histogram has been widely
equalization and expanded. However, the result image has
C. Proposed Histogram Equalization Algorithm certain over enhanced effect. The performance of dividing
In order to overcome the drawback of HE, we presente a histogram into three departments and four departments are
new algorithm to calculate the PDF as follows: presented separately in Fig. 3(e) and Fig. 3(g). There are
corresponding histogram is displayed in Fig. 3(t) and Fig.
nk
perk) =_ k=O, .,L-1
.. (3) 3(h). Fig. 3 proves that as we increase the number of sub­
N layer, the image enhancement has become better and the

VI-347 Volume 1
2010 International Conference On Computer Design And Appliations (ICCDA 2010)

shape of enhanced histogram is more and more adjacent corresponding equalized histogram. However, the shape of
with the original histogram. its histogram is not closed with the original one.
Fig. 4 and Fig. 5 shows another two examples of using Fig.4 Medical image Comparing HE,BHE and Proposed method
HE, BHE and the proposed algorithm respectively on
Medical image and Children image.
Figure 3 Comparing the performance based on different sub-layers

(i) Medical image G) HE

(a) Lena (b) Histogram of(a)

(k) BHE (1) Proposed method


Fig. 5 Children image Comparing HE,BHE and Proposed method
(c) Divide two layers (d) Histogram of(c)

(e) Divide three layers (f) Histogram of(e) (m) Children image (n) Histogram of(m)

(g) Divide four layers (h) Histogram of(g)


Fig. 4 presents our proposed method applying in medical
pictures. Fig. 4 (i) is the original image, in which the bright (0) HE (p) Histogram of(0)
point is a local lump. Though we can see clearly the location
of the lump, there are unable to distinguish its surrounding
muscular tissue. Fig 4 G) shows the performance of HE. The
overall brightness of the image has been improved in certain
degree. However the contrast of image has been reduced.
Fig. 4 (k) displays the result of BHE. A large area of
brightness presents in front of us which demonstrates the
picture has been over-enhanced, so that we can't define the
status of pathogens. Fig. 4 (1) shows the result of proposed (q) BHE (r) Histogram of(q)
algorithm. This method not only successfully enhances the
bright point and many details information but also improves
the contrast of image.
Fig. 5 reveals another example of using HE, BHE and our
proposed method on a low contrast image. Fig. 5 (m) and
Fig. 5 (n) separately show a children picture and its
histogram which is located in a very narrow region. Fig. 5 (0)
and Fig. 5 (P) presents the processed result of HE and (s) Proposed method (t) Histogram of(s)

Fig. 5 (r) describes the histogram which is applied by

Vl-348 Volume 1
2010 International Conference On Computer Design And Appliations (ICCDA 2010)

BHE. The histogram is divided into two departments by [4] A. U. K. Mohammad et ai, "Vessel enhancement using directional
average value. There is some certain degree of compression features," Information Technology Journal, v 6, n 6, pp 851-857,
2007.
in the histogram. Fig. 5 (s) and Fig. 5(t) are the results of
[5] H. Fronthaler, K. Kollreider et ai, "Local features for enhancement
our proposed algorithm. From them, we can prove that the and minutiae extraction in fingerprints," IEEE Trans. on Image
contrast of image is he best one and the shape of histogram Processing,v 17,n 3,pp 354-363,2008.
is more adjacent with the original one. Additionally, the [6] F. Gasparini,S. Corchs et ai ,"Low-quality image enhancement using
distribution of mapping gray level is the densest in all. visual attention," Optical Engineering,v 46,n 4,2007.
From Fig. 3-Fig. 5, they can be seen that the proposed [7] S. C. Pei, Y. C. Zeing, and C. H. Chang, "Virtual restoration of
ancient chinese painting using color contrast enhancement and lacuna
algorithm does good performance in the filed of image texture synthesis," IEEE Trans. Image Processing, vol. 13, pp. 416-
enhancement. 429,2004
[8] W. A., S. H. Chin, and E. C. Tan, "Novel approach to automated
IV. CONCLUSION fingerprint recognition," IEEE Proceedings Vision Image and Signal
This paper proposed a new method for image contrast Processing,vol. 145,pp. 160-166,1998.
enhancement which is especially suitable for multiple-peak [9] A.de la Torre, A. M. Peinado, J. C. Segura et ai, "Histogram
equalization of speech representation for robust speech recognition,"
images. The performance of this algorithm was compared to IEEE Trans. Speech Audio Processing,vol. 13,pp. 355-366,2005.
some classical histogram equalizations. The result [10] S. M. Pizer, "The medical image display and analysis group at the
demonstrates that the proposed algorithm has good university of north carolina: Reminiscences and philosophy,"IEEE
performance in the field of image enhancement. Due to its Trans. Med. Imag.,vol. 22,pp 2-100,2003.
simplicity, it can be realized by simple hardware and [II] S. M. Zhou, J. Q. Gan et al"Interactive image enhancement by fuzzy
consumer electronics. relaxation,"International Journal of Automation and Computing, v 4,
n 3,pp 229-235,2007
REFERENCE [12] B. J. Wang, S. Q. Liu et ai, "A real-time contrast enhancement
[I] J. L. Jiang, Y. S. Zhang et ai, "Local histogram equalization with algorithm for infrared images based on plateau histogram. " Infrared
brightness preservation," Electronic Journal, v 34, n 5, pp 861-866, Physics & Technology,v 48,n I,pp 77-82,2006
Nov. 2006. [13] H. Ibrahim, N. S. PikKong, "Image sharpening using sub-regions
[2] Y. T. Kim, "Quantized bi-histogram equalization," in IEEE Conf. on histogram equalization," IEEE Transactions on Consumer Electronics,
Acoustics, Speech, and Signal Processing , vol. 4, pp 2797-2800, v 55,n 2,pp 891-895, 2009.
1997. [14] A. Ziaei, H. Yeganeh, K. Faez, S. Sargolzaei, "A novel approach for
[3] N. Sengee, H. K. Choi, "Brightness preserving weight clustering contrast enhancement in biomedical images based on histogram
histogram equalization," IEEE Transactions on Consumer Electronics, equalization," New Development and the Future Proceedings of the
v 54,n 3,pp 1329-1337,2008. 1st International Conference on Bio-Medical Engineering and
Informatics,BMEI 2008,v I,pp 855-858

VI-349 Volume I

You might also like