Scholarly article on topic 'Performance Analysis of HE Methods for Low Contrast Images'

Performance Analysis of HE Methods for Low Contrast Images Academic research paper on "Materials engineering"

CC BY-NC-ND
0
0
Share paper
Academic journal
Procedia Computer Science
OECD Field of science
Keywords
{"image enhancement" / "histogram equalization" / RESIHE / MMSICHE / QDHE / PSNR}

Abstract of research paper on Materials engineering, author of scientific article — Subhasmita Sahoo, Jagyanseni Panda, Mihir Narayan Mohanty

Abstract The image enhancement is one of the important issues in image processing. The main purpose is to highlight certain characteristic of image such as: contrast, sharpening. Histogram equalization is the well-known method for image enhancement. Histogram equalization became a popular technique because it is simple and effective. However Histogram equalization cause excessive contrast enhancement which cause visual artifacts of processed image. In this paper new forms of histogram equalization are overviewed to overcome this drawback. The major difference among the methods is the way to divide the input histogram. Recursive exposure based sub-image histogram equalization (R_ESIHE) use average intensity value as the separating point. Median-mean based sub-image clipped histogram equalization (MMSICHE) and Quadrants dynamic histogram equalization for contrast enhancement (QDHE) use median intensity value as separating point. Here objective parameters are Peak signal to noise ratio (PSNR) and Absolute Mean Brightness Error (AMBE)used to compare the quality of enhancement.

Academic research paper on topic "Performance Analysis of HE Methods for Low Contrast Images"

(8)

CrossMark

Available online at www.sciencedirect.com

ScienceDirect

Procedía Computer Science 92 (2016) 72 - 77

2nd International Conference on Intelligent Computing, Communication & Convergence

(ICCC-2016) Srikanta Patnaik, Editor in Chief Conference Organized by Interscience Institute of Management and Technology Bhubaneswar, Odisha, India Performance Analysis of HE Methods for Low Contrast Images Subhasmita Sahoo, Jagyanseni Panda, Mihir Narayan Mohanty*

ITER, SOA University, Bhubaneswar, Odisha, India

Abstract

The image enhancement is one of the important issues in image processing. The main purpose is to highlight certain characteristic of image such as: contrast, sharpening. Histogram equalization is the well-known method for image enhancement. Histogram equalization became a popular technique because it is simple and effective. However Histogram equalization cause excessive contrast enhancement which cause visual artifacts of processed image. In this paper new forms of histogram equalization are overviewed to overcome this drawback. The major difference among the methods is the way to divide the input histogram. Recursive exposure based sub-image histogram equalization (R_ESIHE) use average intensity value as the separating point. Median-mean based sub-image clipped histogram equalization (MMSICHE) and Quadrants dynamic histogram equalization for contrast enhancement (QDHE) use median intensity value as separating point. Here objective parameters are Peak signal to noise ratio (PSNR) and Absolute Mean Brightness Error (AMBE)used to compare the quality of enhancement.

© 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.Org/licenses/by-nc-nd/4.0/).

Peer-reviewunder responsibility oftheOrganizingCommittee of ICCC2016

Keywords:image enhancement, histogram equalization, RESIHE, MMSICHE, QDHE, PSNR

* Corresponding author. Tel.: +91 9437056742; fax: +91 674 2351883. E-mail address:mihir.n.mohanty@gmail.com

1877-0509 © 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.Org/licenses/by-nc-nd/4.0/).

Peer-review under responsibility of the Organizing Committee of ICCC 2016 doi:10.1016/j.procs.2016.07.325

1. Introduction

Now a days use of mobile phones and digital cameras has been tremendously increased as it is very easy to take pictures & light for portability. But sometimes pictures taken by mobile cameras are of low quality due to low light environment or because of any unnatural light sources & surrounding condition. Sometimes by our unconsciousness also mobile phones being shaked and gives blurred image. So many details of the image can't be perceived by normal human eyes. So post processing of the images is necessary. So contrast enhancement and preservation of brightness of the original images plays a very vital role to overcome such problems giving natural enhancement of the image. Thus small details of the image can easily be perceived by human eyes. Histogram equalization (HE) is a very popular method of contrast enhancement technique because of its ease of implementation. HE flattens the probability distribution and stretches the dynamic range of gray level. Hence HE improves the overall contrast of the image and preserves the mean brightness value [1]-[2]. It has application over vast areas such as medical imaging system (X-ray), texture synthesis, video enhancement and speech recognition etc. HE is not suitable for consumer electronics systems as it tends to change the mean brightness of the image to the middle level of the gray level range which produces annoying artifacts such as over enhancement, noise amplification, saturation effect.

Various methods have been proposed by researchers to overcome the drawbacks of traditional histogram equalization method. In 1997 Kim [2] proposed a method namely brightness preserving bi-histogram equalization (BBHE) for brightness preservation and contrast enhancement. This method bisects the histogram into two sub-histograms based on the mean brightness value of the input image and finally equalizes each sub- histogram. In 1999 Wan et al. [3] introduced a method named dualistic sub-image histogram equalization (DSIHE) .This method separates the histogram based on the median value thus containing approximately equal number of pixel. Wan et al. claimed that DSIHE method is better than BBHE brightness preservation and entropy of the image. Chen and Ramli proposed a method named minimum mean brightness error bi- histogram equalization (MMBEBHE). This method can preserve the mean brightness optimally [4]. MMBEBHE iteratively calculate absolute mean brightness error (AMBE) and bisect the histogram into two sub- histogram with minimum AMBE. Chen and Ramli has also proposed another method named recursive mean separate histogram equalization (RMSHE) [5] which is another version of BBHE. RMSHE iteratively performs BBHE. This method recursively separates the histogram into two sub-histogram based on the average brightness value of the input image and after sub division BBHE is performed to each sub-histogram. The algorithm is performed recursively until the desired number of sub-histogram is found. A similar technique is proposed by Sim et al.[6] named recursive sub-image histogram equalization(RSIHE). In this method sub-division is based on median value of brightness value of the input image. But no significant enhancement result is found for iteration to find the optimal value.

Kim and Chung [7] has introduced a method named recursively separated and weighted histogram equalization (RSWHE). This method divides the histogram iteratively and modifies the histogram by means of a weighting process using normalized power law function. But these techniques do not provide the mechanism for enhancement rate. In [8-10] Singh and Kapoor have proposed Exposure based Sub Image Histogram Equalization (ESIHE) method for enhancement of low exposure images where the exposure threshold value is being used for decomposition of histogram. The histogram equalization technique is the most popular technique and is applied in many ways by number of researchers.

This paper is organized as follows: section 2 describes different contrast enhancement methods, section 3 provides experimental results and section 4 concludes the paper.

2. Proposed Method

There are different types of contrast enhancement techniques applied by many authorsas described in literature. But in this work, authors tried on histogram based approach and compared among the proposed techniques. Though it is a common technique, still its variants work different way and made it attractive. Some of them are discussed in this section.

Histogram equalization (HE)

The traditional histogram equalization method can be described as follow:

Consider a input image I. I={I (i, j)} is an image having L discrete gray levels of the range [0 K-1]. H (I) is the histogram of the image. The probability density function (PDF) of the image can be defined as

pdf(k) = for k=0,1,.....L-l (1)

Where N is the total number of pixels.

From the PDF the cumulative distribution function (CDF) can be defined as

cdf (k) = £ pdf ( j)for k=0,1,.....L-1 (2)

CDF is used for mapping the input image having gray level Xk into output gray level fk is defined as fk =h+(IL.x-h)-cdf(k) (3)

Hence, HE remaps the input image into the entire dynamic range [/0, IL_X ].

2.1. Recursive exposure based sub-image histogram equalization (R-ESIHE)

This method [11] recursively performs ESIHE and stops when exposure difference value among successive iteration is less than a threshold value which is predefined. This section presents the algorithm of RESIHE. The algorithm consists of three steps such as: 1) exposure threshold calculation, 2) histogram clipping and 3) histogram sub division and equalization.

2.1.1. Exposure threshold calculation This method is helpful for dividing the image into low or high exposure. Exposure value ranges from [0-1]. If the exposure value is less than 0.5 and greater than 0.5 then the image contains majority of low exposure region and over exposure region respectively.

lJ^h (k )k

Exp= ^--(4)

Where H (k) is histogram of image and L is total number of gray levels. Exposure threshold is given by

I expm=L(1-Exp) (5)

2.1.2. Histogram clipping

For limiting the degree of enhancement rate we can limit the histogram in order to overcome over enhancement. Clipping threshold is calculated by average of the number of intensity. The histogram bins having higher value than threshold value are replaced by the threshold value.

Thdip = mean[H(k)] (6)

Hclip(k) = Thclip for h(k)>Tclip (7)

H (k) and Hclip (k) are the original and clipped histogram respectively.

2.1.3. Histogram sub division and equalization

The original histogram is bisected into two sub-images RL and Ru ranging from gray level 0 to IExpTh -1 and

IExpTh to L-1. PDFl (k) and PDFU (k) are corresponding PDF of the sub images. CDFL and CDFU are corresponding CDFs are being calculated. Then the individual histograms are equalized by using the transfer

function. These are used for equalizing the sub histograms individually. Then integration of both the sub image is done to get a complete image. The number of iteration is decided based on the threshold OC whose value is normally taken very less. The number of recursion is dependent on the successive iterations. If the exposure difference between successive iteration is less than threshold OC (0.1) then the iteration stops.

2.2.4 Algorithm for R-ESIHE

Step 1. Histogram H(k) of the original image is calculated. Step 2. Exposure value Exp and threshold IExpTh are calculated.

Step 3. Clipping threshold Thclip and Hclip (k) are calculated.

Step 4. Then sub-histograms are equalized individually.

Step 5. Combine the images into one image.

Step 6. If the exposure difference between successive iteration is less than the threshold (OC=0A) then we will get the output image otherwise step 1 -5 are being repeated.

2.2. Median-mean based sub-image clipped histogram equalization (MMSICHE)

The proposed algorithm [9] consists of three steps Such as: 1) Calculation of median and mean values, 2) Clipping of histogram using a plateau limit, 3) Sub division and equalization of histogram.

2.2.3. Sub division and equalization of histogram The original histogram is first bisected based on median intensity value M .These individual sub-histograms are further decomposed into two smaller sub histograms based on individual mean intensity values Mmland Mmu

.Thus total four number of sub images are formed Ru, RLU, RUl, RUu ranging from gray levels 0 to Mml, Mml+1 to

M, M+1 to Mmu, Mmu+1to L-1. PDFli (k), PDFlu(k), PDFui (k), PDFuu (k) are the corresponding PDFs and can

be calculated. The next step of MMSICHE is to equalize the four sub histograms individually by using the transfer function. These are transfer functions used for equalizing the sub histograms individually. Then integration of the four sub image is done to get a complete image.

2.3.4 Algorithm for MMSICHE

Step 1. Histogram H(k) of the original image is calculated.

Step 2. Median value M and mean values Mml and Mml are calculated.

Step 3. Clipping threshold Thclip and Hclip (k) are calculated.

Step 4. Then sub-histograms are equalized individually. Step 5. Combine the images into one image.

2.3. Quadrants dynamic histogram equalization for contrast enhancement (QDHE)

QDHE algorithm [11] separates the histogram into four quadrants. Then the resultant sub histograms are clipped according to the mean value of input image. Each sub histogram doesn't ensure the balance space. So sufficient contrast enhancement may not occur. Thus a new dynamic gray level range is assigned to each sub histogram. After that each sub histogram is equalized and combined to form a complete image. QDHE algorithm divides the histogram into four sub images based on the median value of the image. The median based partition approach tends to segment the number of pixels equally in each sub histogram.

For the total number of pixels in the histogram of the input image m1, m2 and m3 are intensities set at 0.25, 0.50 and 0.75 respectively.

2.4.1. Algorithm for QDHE

Step 1. Histogram H (k) of the original image is calculated. Step 2. Median values m1, m2 and m3 are calculated.

Step 3. Clipping threshold ThcMp and clipped histogram Hcllp(k) are calculated.

Step 4. New gray level dynamic range is allocated. Step 5. Then sub-histograms are equalized individually. Step 6. Combine the images into one image.

3. Results and discussion

In this section, the performance of four different methods of contrast enhancement techniques i.e. HE, MMSICHE, QDH and R-ESIHE are being analysed. In order to analyze and compare these four methods five images namely moon, coins, AT3_1m4_01, cell and rice are taken to test the accuracy of the proposed methods of contrast enhancement. From Fig. 1 through Fig. 5, it is clearly observed that the R-ESIHE increases the contrast of the original image with natural appearance without intensity saturation and noise amplification and controls the degree of enhancement. Simultaneously, peak signal to noise ratio (PSNR) and absolute mean brightness error (AMBE) is calculated and evaluated to test the robustness of the proposed method. Results of the comparison are summarized in Table 1 and Table 2. From these results it is clearly observed that in comparison to other methods, R-ESIHE produces images having high PSNR value thus reduces noise components in the enhanced image. Similarly R-ESIHE is having less value of AMBE which means this method preserves the mean brightness value effectively in comparison to other methods.

lb ^^n^^Hc I

Fig. 1. Simulation result of 'moon.tif (a) original image, (b) HE, (c) QDHE, (d) MMSICHE, (e) R-ESIHE

■a ^HI^^H) " <SHc d Hi^^He

Fig. 2. Simulation result of ■coins.png' lal original image, lb I HE, Id QDHE, Id) MMSICHE, (el R-ESIHE

Fig. 3. Simulation result of 'AT3_1m4_01.tif (a) original image, (b) HE, (c) QDHE, (d) MMSICHE, (e) R-ESIHE

« WtBf.ni^BM DBH

498 II R "

ISHSlSb d

lt of 'cell.tif (a) original image, (b) HE, (c) QDHE, (d) MMSICHE, (e) R-ESIHE

Fig. 4. Simulatk

ib ^b^Hc i

Fig. 5. Simulation result of 'rice.png' (a) original image, (b) HE, (c) QDHE, (d) MMSICHE, (e) R-ESIHE

Table 1. PSNR

Images HE MMSICHE QDHE R-ESIHE

Moon.tif 9.3805 27.0861 47.7695 26.9739

Coins.png 15.2092 22.3301 25.5029 23.2925

AT3_1m4_01.tif 10.3674 29.3177 30.4564 21.2761

Cell.tif 11.6303 22.7983 30.3639 19.3113

Rice.png 16.2825 24.7479 28.6016 21.1397

Average 13.9401 24.5205 33.6362 23.1720

Table 2. AMBE

Images HE MMSICHE QDHE R-ESIHE

Moon.tif 74.3539 3.7187e+003 3.7359 1.6187

Coins.png 24.3703 2.3391e+003 30.7527 6.2162

AT3_1m4_01.tif 50.9241 3.3492e+003 35.7932 5.2474

Cell.tif 9.7254 1.3435e+003 16.9107 21.2617

Rice.png 16.1298 2.7139e+003 2.3901 5.7617

Average 21.7623 2.6844e+003 17.4278 7.7781

4. Conclusion

In this paper, five different contrast enhancement techniques has been proposed and compared to test the accuracy, quality and effectiveness. And simulation result clearly show that R-ESIHE method not only preserves the brightness of the image but also controls rate of enhancement or over enhancement In this method clipping of histogram provides control on degree of enhancement. Thus R-ESIHE produces superior image in comparison to other methods.

References

1. R.C. Gonzalez, R.E. Woods, Digital Image Processing, 2nd ed., Prentice-Hall, Englewood Cliffs, NJ,2002.

2. Y.T. Kim, Contrast enhancement using brightness preserving bi-histogram equalization, IEEE Trans. Consum. Electron. 43 (1997)1-8.

3. Y.Wan, Q. Chen, B.M.Zhang, Image enhancement based on equal area dualistic sub-image histogram equalization method, IEEE Trans. Consum. Electron. 45 (1999) 68-75.

4. S.D.Chen, A.R.Ramli, Minimum mean brightness error bi-histogram equalization in Contrast enhancement, IEEE Trans. Consum. Electron. 45 (2003) 1310-1319.

5. S.D.Chen, A.R.Ramli, Contrast enhancement using recursive mean-separate histogram equalization for scalable brightness preservation, IEEE Trans. Consum. Electron. 49 (2003) 1301-1309.

6. K.S.Sim, C.P.Tso, Y.Y.Tan, Recursive sub-image histogram equalization applied to gray scale image, Pattern Recogn. Lett.28 (2007) 1209-1221.

7. M.Kim, M.G.Chung, Recursively separated and weighted histogram equalization for brightness preservation and Contrast enhancement, IEEE Trans. Consum. Electron. 54 (2008) 1389-1397.

8. K.Singh, R.Kapoor, Image enhancement using exposure based sub-image histogram equalization, Pattern Recogn. Lett. 36(2014) 1014.

9. K.Singh, R.Kapoor, Image enhancement via Median-Mean Based Sub-Image Clipped histogram equalization, Pattern Recogn. Lett. 125(2014) 4646-4650.

10. K.Singh, R.Kapoor, Enhancement of low exposure images via recursive histogram equalization algorithms, Pattern Recogn. Lett. (2015) (In Press).

11. C.H.Ooi, N.A.M.Isa, Quadrants Dynamic Histogram Equalization for Contrast Enhancement, IEEE Trans. Consum. Electron. 56 (2010)2552-2559.

12. T.L.Tan, K.S.sim, C.P.Tso, Image enhancement using background brightness preserving histogram equalization,Electron. Lett.48(2012) 155-157.