Image contrast enhancement for preserving mean brightness without losing image features

Engineering Applications of Artificial Intelligence 26 (2013) 1487–1492

Contents lists available at SciVerse ScienceDirect

Engineering Applications of Artificial Intelligence journal homepage: www.elsevier.com/locate/engappai

Image contrast enhancement for preserving mean brightness without losing image features Shih-Chia Huang n, Chien-Hui Yeh Department of Electronic Engineering, National Taipei University of Technology, Taiwan

a r t i c l e i n f o

abstract

Article history: Received 15 July 2012 Received in revised form 6 September 2012 Accepted 29 November 2012 Available online 29 January 2013

Histogram equalization is a well-known and effective technique for improving the contrast of images. However, the traditional histogram equalization (HE) method usually results in extreme contrast enhancement, which causes an unnatural look and visual artifacts of the processed image. In this paper, we propose a novel histogram equalization method that is composed of an automatic histogram separation module and an intensity transformation module. First, the proposed histogram separation module is a combination of the proposed prompt multiple thresholding procedure and an optimum peak signal-to-noise ratio (PSNR) calculation to separate the histogram in small-scale detail. As the final step of the proposed process, the use of the intensity transformation module can enhance the image with complete brightness preservation for each generated sub-histogram. Experimental results show that the proposed method not only retains the shape features of the original histogram but also enhances the contrast effectively. & 2012 Elsevier Ltd. All rights reserved.

Keywords: Contrast enhancement Histogram equalization Image quality evaluation

1. Introduction Over the last few decades, visual image quality has been actively improved using contrast enhancement techniques, which are increasingly necessary for the design of digital multimedia systems, such as video surveillance systems (Havasi et al., 2007; Eng et al., 2008), digital photography (Oakley and Bu, 2007; Fattal, 2008), medical imaging systems (Chen et al., 2005; Yang et al., 2009), and low power systems (Lai and Tsai, 2008). Moreover, contrast enhancement has become widely available to provide a ‘‘better’’ transform representation for real image processing systems with both software and hardware environments, including Photoshop (Chen et al., 2006a), mobile devices, digital TV, and digital cameras (Kim et al., 2001; Kim, 1997; Wang et al., 1999; Sim et al., 2007). In recent years, contrast enhancement techniques have been used in many digital multimedia systems, such as for object tracking in video surveillance systems (Chen et al., 2008), radiography applications (Dippel et al., 2002), arterial visualization of medical imaging (Bemmel et al., 2003), tumor microcirculation (Koh et al., 2003), virtual restoration of ancient Chinese paintings (Pei et al., 2004), vision impairment estimation (Tang et al., 2004), recovery of underwater visibility (Schechner and Karpel, 2005), bas-relief generation (Sun et al., 2009), and face recognition (Xie and Lam, 2005). Therefore, the aim of this study is to develop a

n

Corresponding author. E-mail address: [email protected] (S.-C. Huang).

0952-1976/$ - see front matter & 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.engappai.2012.11.011

high-performance contrast enhancement method that fits the current requirements (Perner et al., 1999; Shin et al., 1992; Zhang et al., 2010). Contrast enhancement methods can be broadly categorized into two major classes: direct and indirect methods (Arici et al., 2009). Direct methods (Beghdadi and Negrate, 1989; Cheng and Xu, 2000) try to improve image using the definition of a contrast measure. Indirect methods (Sherrier and Johnson, 1987; Polesel et al., 2000) exploit the under-utilized dynamic range of the images to improve contrast without a contrast measurement. Almost all the most popular contrast enhancement methods in the literature fall into the second category (Arici et al., 2009). Furthermore, the indirect methods can be divided into two subclasses: histogram modification techniques (i.e., spatial domain techniques) and transform domain techniques (Arici et al., 2009; Hanmandlu and Jha, 2006). In particular, histogram modification techniques have attracted significant attention from researchers due to their simplicity and computational efficiency. Generally speaking, histogram modification techniques can be categorized into two main types, global histogram modification and local histogram modification (Kim et al., 2001; Lamberti et al., 2006). Global histogram modification techniques attempt to modify the spatial histogram of an image in order to closely match a uniform distribution via the transform function. This is generated by using the histogram information of the entire input image (Abdullah-Al-Wadud et al., 2007). However, global histogram modification technique cannot adapt to local brightness features. This causes limitations in the amount of contrast enhancement in

1488

S.-C. Huang, C.-H. Yeh / Engineering Applications of Artificial Intelligence 26 (2013) 1487–1492

some parts of the image (Chen et al., 2006a; Abdullah-Al-Wadud et al., 2007). A direct extension of global histogram modification is termed local histogram modification. In order to efficiently improve the visibility of the small-scale detail, local histogram modification techniques independently equalize each sub-histogram on the basis of histogram separation (Chen et al., 2006b). However, local histogram modification techniques sometimes cause overenhancement and noise-enhancing artifacts in some portion of an image (Abdullah-Al-Wadud et al., 2007). A good contrast enhancement technique should specifically address several significant properties, some of which are listed below. (1) Noise tolerance: The contrast enhancement technique should exhibit appropriate noise immunity. (2) Uniform contrast: The contrast enhancement technique should provide uniform contrast of the entire image. (3) Brightness preservation: The contrast enhancement technique should enhance the contrast of the image without losing brightness. (4) Convenient implementation: The contrast enhancement technique should be able to be set up quickly and reliably. In this paper, a novel contrast enhancement method is presented to enhance the contrast of an image without losing the original histogram characteristics. It is expected to eliminate the abovementioned drawbacks of the conventional global-based and local-based histogram modification methods effectively. The proposed method can be briefly described as follows: 1. Separation of a histogram with the proposed multiple thresholding procedure by using a prompt mean function and standard deviation. 2. Achievement of contrast enhancement by equalizing subhistograms in small-scale detail. Experimental results show that the proposed method gives more accurate results than other state-of-the-art methods in the case of a wide range of natural digital images. The performance of the proposed method will be demonstrated through qualitative and quantitative evaluations. The rest of this paper is divided into Sections 2–5. In Section 2, we present a relatively compact overview of the popular histogram modification methods. Section 3 describes the proposed contrast enhancement method in detail. Section 4 presents a comparison of the experimental results of the proposed method and some of the other existing methods. Our concluding remarks are presented in Section 5.

2. Related work Known as one of the most popular contrast enhancement techniques, histogram equalization (HE) can efficiently perform contrast enhancement because of its simplicity and effectiveness. The basic idea of the HE method is to re-map the gray levels of an input image using a transformation function with the cumulative distribution of the input image. HE attends to and stretches the dynamic range of the image histogram to improve the overall contrast of the original image. However, the HE method is unsuitable for consumer electronic applications because the calculated transformation function may significantly change the brightness of the original input image. To overcome this problem, several researchers have studied the preservation of image brightness in HE-based enhancement

methods. In 1997, brightness preserving bi-histogram equalization (BBHE) (Kim, 1997) was proposed for separating an image histogram into two sub-histograms with the mean of the graylevel for the input image before equalizing them independently. Later, Wang et al. (1999) proposed dualistic sub-image histogram equalization (DSIHE) for separating a histogram on the basis of the median instead of the mean of the gray levels. Sim et al. (2007) proposed recursive sub-image histogram equalization (RSIHE) by extending DSIHE (Wang et al., 1999). In RSIHE (Sim et al., 2007), the median-based histogram separation is applied several times to obtain the local median values, whereas DSIHE (Wang et al., 1999) performs the separation only once. Although the HE method can attain a significant improvement in image contrast, the equalized image may distort the mean brightness of the input image (Kim, 1997). Because the BBHE (Kim, 1997) method only performs the local histogram equalization with one separation, the enhanced image may contain an extreme over-enhancement and noise artifacts. Further, because the DSIHE method (Wang et al., 1999) is similar to the BBHE method (Kim, 1997) with respect to a single histogram separation, an image enhanced using DSIHE may contain an extreme over-enhancement and noise artifacts. Although the RSIHE method (Sim et al., 2007) is better than the traditional HE, BBHE (Kim, 1997), and DSIHE (Wang et al., 1999) methods, some serious over-enhancement and noise artifacts still exist in the enhanced image because of the number of sub-histograms generated experimentally using the calculated local median values. As mentioned before, the number of sub-images and its corresponding calculation of threshold determination necessitate the development of an effective histogram equalization algorithm.

3. Proposed method In this section, we propose a novel contrast enhancement method, which involves two important modules: a histogram separation module and an intensity transformation module. First, the proposed histogram separation module is a combination of the proposed prompt multiple thresholding procedure and an optimum PSNR calculation to separate the histogram in small-scale detail. As the final step of the proposed process, the use of the intensity transformation module can enhance the image with complete brightness preservation for each generated sub-histogram.

3.1. Histogram separation Each M  N discrete input image Ixy can be defined as an M-byN matrix, where xy denotes an arbitrary pixel in the spatial domain. The matrix of the original image Ixy is expressed as follows: 2 3 I0,0 I0,1    I0,N1 6 7 I1,1    I1,N1 7 6 I1,0 7: Ixy ¼ 6 ð1Þ 6 ^ 7 ^ & ^ 4 5 IM1,0 I0,1    IM1,N1 (1) Prompt multiple threshold selection: Suppose that an input image I is composed of G discrete intensity levels; thus, the statistical histogram H of an input image I is expressed as follows: HðhÞ ¼ nh

where h ¼ 0,1,2, . . . ,G1:

ð2Þ

Note that nh represents the number of pixels that correspond to the intensity h of the input image I. Based on the statistical histogram H of the input image I, the mean value m can be

S.-C. Huang, C.-H. Yeh / Engineering Applications of Artificial Intelligence 26 (2013) 1487–1492

calculated using the following mean function: Pb h ¼ a h  HðhÞ : m¼ P b h ¼ a HðhÞ

is lower than a particular value of 0.1, which tends to saturate the sub-histograms generation by separating the image histogram. ð3Þ 3.2. Intensity transformation

After the mean value m is calculated, the standard deviation function s is expressed as follows: !1=2 Pb 2 h ¼ a ðhmÞ  HðhÞ s¼ : ð4Þ Pb h ¼ a HðhÞ Notice that ½a,b represents the interval of calculation, which is initialized as ½0,255. According to the calculated mean value and standard deviation, three threshold values, ms, m, and m þ s, can be defined to further separate the histogram. (2) Optimum recursion level: For each recursion level, the low boundary is set to ms and the high boundary b is set to m þ s. According to the new calculated interval ½a,b, mean function, and standard deviation, the two new threshold values ms and m þ s are calculated as an addition to the thresholds, and then, the existing threshold is replaced with m when it is closest to the new mean value. In order to accurately decide the appropriate number of subimages, we calculate the peak signal-to-noise ratio (PSNR) (Arora et al., 2008) at each recursion level. The PSNR is defined as follows: PSNR ¼ 20 log10

255 , RMSE

ð5Þ

where RMSE is the root-mean-square error, defined as !1=2 1 N 1 X X 1 M 2 ^ RMSE ¼ ½I xy Ixy  : MN x ¼ 0 y ¼ 0

1489

Based on the histogram separation, the optimum number t of sub-images can be expressed as follows: Ikxy ¼ fIxy 9Sk rIxy o Sk þ 1 ,8Ixy A Ig,

ð7Þ

where k ¼ 0,1,2, . . . ,t1, and Itxy represents each sub-image. (1) Probability density function and cumulative distribution function: For each sub-image Ik, the probability density function (pdf) is expressed as follows: n k pdf ðhÞ ¼ P h , nh

ð8Þ k

where h ¼ Sk þ1,Sk þ 2, . . . ,Sk þ 1 . Notice that pdf ðhÞ is associated with the histogram of the k-th sub-image to represent the frequency of a specific input intensity h. Then, the cumulative distribution function (cdf) is expressed as follows: k

cdf ðhÞ ¼

h X

k

ðpdf ðeÞÞ,

ð9Þ

k

e ¼ S þ1

(2) Transformation mapping function: Finally, the transformation mapping function is used for enhancing the original image. This is characterized by utilizing the cdfk of sub-image Ik for k segments. Then, the transformation mapping function is defined as follows: k

T k ðhÞ ¼ Sk þ ðSk þ 1 Sk Þ  cdf ðhÞ,

ð10Þ k

ð6Þ

Here, I and I^ are the original and enhanced images of size M  N, respectively. Note that a higher PSNR value represents a better image quality. According to Arora et al. (2008), the optimum recursion level can be determined when the increasable PSNR

k

kþ1

where k ¼ 0,1, . . . ,t1 and h ¼ S þ 1,S þ 2, . . . ,S . The flowchart of the proposed approach involves an automatic histogram separation module and an intensity transformation module for achieving our objective of contrast enhancement, as shown in Fig. 1. As shown in the top-left of Fig. 1, the original image has insufficient contrast between the human face and the background. In order to enhance the contrast along with complete

Prompt multiple threshold selection Increasable PSNR < 0.1?

PSNR calculation

Traditional HE Proposed method

Fig. 1. Flowchart of the proposed contrast enhancement method.

1490

S.-C. Huang, C.-H. Yeh / Engineering Applications of Artificial Intelligence 26 (2013) 1487–1492

brightness preservation, the automatic histogram separation module separates the original histogram into small-scale detail. Each separated sub-histogram region is shown in the top-right of Fig. 1. Based on the separated sub-histograms, the piecewise transformation function illustrates the hybrid mapping curve shown in the bottom-right of Fig. 1. The solid line represents the mapping curve of the proposed method, and the dotted line represents the mapping curve of traditional HE. As can be seen, the proposed method has a smoother curve than the traditional HE. Notice that the x-coordinates represent the original intensity of the input image, and the y-coordinates denote the enhanced intensity after the contrast enhancement is achieved. As a result, the enhanced image mapped by using the proposed smooth mapping curve produces a relatively perceivable contrast between human hair, face, and the background, as shown in the bottom-left of Fig. 1.

4. Experimental results For a robust visual quality of contrast enhancement, some specific properties are extremely desirable in an effective contrast enhancement method. These are the noise tolerance ability, brightness preservation of uniform contrast enhancement, and convenient implementation. In order to demonstrate the advantages of the proposed method for the necessary specific properties, this section presents a comparison of the proposed method with some of the aforementioned existing methods. Experimental results are produced using a variety of original images in three evaluations: 1. Qualitative evaluation of enhancement results is carried out using each method.

2. Quantitative evaluation for all test images is carried out on the basis of two well-known metrics. 3. Time complexity is evaluated for each contrast enhancement method.

4.1. Qualitative evaluation Fig. 2 shows the original ‘‘Girl’’ image (Fig. 2(a)) and the corresponding contrast-enhanced versions (Fig. 2(b–f)). Fig. 2(b) shows the traditional histogram equalized image of the original ‘‘Girl’’ image. An extreme contrast of the image can be observed between the background and the human hair, body, and face. BBHE (Kim, 1997), DSIHE (Wang et al., 1999), and RSIHE (Sim et al., 2007) use the local histogram information to reduce the extreme effect of HE. However, their histogram-equalized images still have some aspects of the HE result: the hair and clothing are relatively dark; the bright face is relatively bright; and the background region is involved with the amplified noise. This situation can be easily observed in the images of Fig. 2(c–e). Fig. 2(f) shows the enhancement result of the proposed method. With complete brightness preservation, the proposed method gives a relatively natural brightness improvement on the hair and face of the circled region. 4.2. Quantitative evaluation In addition to the qualitative evaluation, accuracy measurement is necessary for a further comparison between the proposed method and other state-of-the-art methods, on the basis of absolute mean brightness error (AMBE) (Kim and Chung, 2008) instead of PSNR (Arora et al., 2008). The performance of brightness preservation is rated using an objective measurement referred

Fig. 2. Digital image ‘‘Girl’’: (a) is the original image; the remaining five images are equalized images generated using (b) the traditional HE method, (c) the BBHE method (Kim, 1997), (d) the DSIHE method (Wang et al., 1999), (e) the RSIHE method (Sim et al., 2007), and (f) the proposed method.

S.-C. Huang, C.-H. Yeh / Engineering Applications of Artificial Intelligence 26 (2013) 1487–1492

to as AMBE, which is defined as the absolute differential gray-level mean between the original image and the enhanced image: AMBE ¼ 9Ie I^e 9:

ð11Þ

Ie and I^e denote the gray-level mean of the original and the enhanced image, respectively. Note that a lower AMBE value indicates better brightness preservation. Table 1 lists the PSNR values of the enhancement results obtained using each method for 15 test images. Because the proposed method enhances the image by equalizing the optimum sub-histograms in small-scale detail, the enhanced images obtained using the proposed method produce neither noise artifacts nor over-enhancement to achieve the highest PSNR values. Table 2 lists the AMBE values of the enhancement results obtained using each method for 15 test images. The proposed method can preserve the complete mean brightness, which is achieved by enhancing the image with the optimum subhistogram equalization. Therefore, the proposed method effectively achieves the lowest AMBE values among all. 4.3. Time complexity Based on the Bachmann–Landau notation, the time complexities of the HE, BBHE (Kim, 1997), DSIHE (Wang et al., 1999), RSIHE (Sim et al., 2007), and the proposed methods are analyzed for the image size M  N. In the case of the traditional HE method, the computation of the histogram requires O(MN) time. Calculating the mapping function from the histogram requires O(L) time, where L denotes the discrete level of the input image. Finally, the enhanced image is obtained using the mapping function in O(MN) Table 1 Comparison of PSNR values for all considered contrast enhancement methods. Test images

HE

BBHE

DSIHE

RSIHE

Ours

Woman Plane Lena Jet Boats City Couple Girl Vacas Carney Donna Cat Peppers Elaine

17.8270 10.0494 16.6746 11.9202 17.9821 15.7870 16.4238 13.0308 16.3450 6.02270 14.3606 15.4340 19.2244 18.6226

17.7926 14.9109 19.6195 20.6928 18.0720 16.8652 16.3811 13.3493 18.1387 15.3223 16.2868 19.1703 19.7440 18.7900

18.3150 13.2561 19.3112 16.0345 18.0801 16.8652 16.4481 13.0274 18.2614 11.2570 16.8542 18.5818 19.7314 18.8154

22.6265 13.4218 24.4033 16.6083 21.7637 16.9494 16.4481 13.0458 23.4952 12.8203 21.2863 18.5818 19.8540 23.7119

31.3312 29.8202 29.9760 29.8388 33.9337 33.4164 35.6083 32.7660 33.8939 26.4214 33.0070 31.6712 30.1995 30.6834

Table 2 Comparison of AMBE values for all considered contrast enhancement methods. Test images

HE

BBHE

DSIHE

RSIHE

Ours

Woman Plane Lena Jet Boats City Couple Girl Vacas Carney Donna Cat Peppers Elaine

15.3867 63.8809 29.6081 49.7900 2.01950 23.9070 5.36790 5.24750 27.2247 122.667 41.1844 33.2338 12.7117 8.1194

15.6674 16.9610 12.7515 0.61260 18.8947 12.8508 7.97810 23.2249 14.8441 26.9072 27.0488 11.4700 5.8611 4.9531

11.0657 27.9712 13.6411 16.1451 10.1940 12.1678 4.61740 7.82180 14.0099 45.6341 23.6191 13.9341 5.9397 4.9531

9.53630 33.9303 9.40430 23.0606 4.83150 12.1678 4.61740 7.94700 14.0099 38.3371 21.1089 13.9341 5.6704 8.1518

0.3579 1.3694 2.6339 1.1781 0.1559 0.5361 0.7159 0.1760 1.5754 3.7902 1.2225 1.3403 1.6923 0.2675

1491

time. Hence, the total time complexity of the traditional HE method is Oð2MN þLÞ. In the case of the BBHE (Kim, 1997) method, the computation of the histogram requires O(MN) time. The computation of the mean value requires O(L) time, where L denotes the discrete level of the input image. Calculating the mapping function from the histogram requires O(L) time. Finally, the enhanced image is obtained using the mapping function in O(MN) time. Hence, the total time complexity of the traditional BBHE (Kim, 1997) method is Oð2MN þ 2LÞ. In the case of the DSIHE (Wang et al., 1999) method, the computation of the histogram requires O(MN) time. The computation of the median value requires O(L) time, where L denotes the discrete level of the input image. Calculating the mapping function from the histogram requires O(L) time. Finally, the enhanced image is obtained using the mapping function in O(MN) time. Hence, the total time complexity of the traditional DSIHE (Wang et al., 1999) method is Oð2MN þ2LÞ. In the case of RSIHE (Sim et al., 2007) method, the computation of the histogram requires O(MN) time. The computation of all the sub-median values requires O(L) time, where L denotes the discrete level of the input image. Calculating the mapping function from the histogram requires O(L) time. Finally, the enhanced image is obtained using the mapping function in O(MN) time. Hence, the total time complexity of the RSIHE (Sim et al., 2007) method is Oð2MN þ2LÞ. In the case of the proposed method, the computation of the histogram requires O(MN) time. Separating the histogram by using mean and standard deviation requires O(L) time, where L denotes the discrete level of the input image. The calculation of the optimum recursive level requires O(L) time. Calculating the mapping function from the histogram requires O(L) time. Finally, the enhanced image is obtained using the mapping function in O(MN) time. Hence, the total time complexity of the proposed method is Oð2MN þ3LÞ. Because the image size M  N is considerably greater than L, the time complexity of each method is rounded off to O(MN). As a result, the proposed method and other state-of-the-art methods are easy to implement for real-time applications.

5. Conclusion In this paper, we presented a novel histogram equalization method for enhancing the contrast of an image. First, the original histogram was separated into several sub-histograms automatically by using the proposed histogram separation module. The generated sub-histograms were then equalized by the intensity transformation module to achieve an accurate contrast enhancement. Experimental results revealed that the proposed method generated very high-quality enhancement images, as demonstrated using a qualitative visual inspection and quantitative PSNR and AMBE evaluations, as compared to the other stateof-the-art methods.

Acknowledgments The authors would like to thank the National Science Council of the Republic of China, Taiwan, for financially supporting this research under Contract no. NSC 100-2628-E-027-012-MY3. References Abdullah-Al-Wadud, M., Kabir, M.H., Dewan, M.A.A., Chae, O., 2007. A dynamic histogram equalization for image contrast enhancement. IEEE Trans. Consum. Electron. 53 (May (2)), 593–600.

1492

S.-C. Huang, C.-H. Yeh / Engineering Applications of Artificial Intelligence 26 (2013) 1487–1492

Arici, T., Dikbas, S., Altunbasak, Y., 2009. A histogram modification framework and its application for image contrast enhancement. IEEE Trans. Image Process. 18 (September (9)), 1921–1935. Arora, S., Acharya, J., Verma, A., Panigrahi, P.K., 2008. Multilevel thresholding for image segmentation through a fast statistical recursive algorithm. Pattern Recognition Lett. 29, 119–125. Beghdadi, A., Negrate, A.L., 1989. Contrast enhancement technique based on local detection of edges. Comput. Vis. Graph. Image Process. 46 (May (2)), 162–174. Bemmel, C.M., Wink, O., Verdonck, B., Viergever, M.A., Niessen, W.J., 2003. Blood pool contrast-enhanced MRA: improved arterial visualization in the steady state. IEEE Trans. Med. Imaging 22 (May (5)), 645–652. Chen, D.R., Chang, R.F., Chen, C.-J., Ho, M.F., Kuo, S.J., Chen, S.T., Hung, S.J., Moon, W.K., 2005. Classification of breast ultrasound images using fractal feature. Clin. Imaging 29 (July–August (4)), 235–245. Chen, Z.Y., Abidi, B.R., Page, D.L., Abidi, M.A., 2006a. Graylevel grouping (GLG): an automatic method for optimized image contrast enhancement-part. ii: the variations. IEEE Trans. Image Process. 15 (August (8)), 2303–2314. Chen, Z.Y., Abidi, B.R., Page, D.L., Abidi, M.A., 2006b. Gray-level grouping (GLG): an automatic method for optimized image contrast enhancement—part i: the basic method. IEEE Trans. Image Process. 15 (August (8)), 2290–2302. Chen, C.Y., Lin, T.M., Wolf, W.H., 2008. A visible/infrared fusion algorithm for distributed smart cameras. IEEE J. Sel. Top. Signal Process. 2 (August (4)), 514–525. Cheng, H.D., Xu, H.J., 2000. A novel fuzzy logic approach to contrast enhancement. Pattern Recognition 33 (5), 809–819. Dippel, S., Stahl, M., Wiemker, R., Blaffert, T., 2002. Multiscale contrast enhancement for radiographies: Laplacian pyramid versus fast wavelet transform. IEEE Trans. Med. Imaging 21 (April (4)), 343–353. Eng, H.L., Toh, K.A., Yau, W.Y., Wang, J., 2008. DEWS: A live visual surveillance system for early drowning detection at pool. IEEE Trans. Circuits Syst. Video Technol. 18 (February (2)), 196–210. Fattal, R., 2008. Single image dehazing. ACM Trans. Graph. 27, 72:1–72:9. Hanmandlu, M., Jha, D., 2006. An optimal fuzzy system for color image enhancement. IEEE Trans. Image Process. 15 (10), 2956–2966. Havasi, L., Szlavik, Z., Sziranyi, T., 2007. Detection of gait characteristics for scene registration in video surveillance system. IEEE Trans. Image Process. 16 (February (2)), 503–510. Kim, Y., 1997. Contrast enhancement using brightness preserving bihistogram equalization. IEEE Trans. Consum. Electron. 43 (February (1)), 1–8. Kim, M., Chung, M.G., 2008. Recursively separated and weighted histogram equalization for brightness preservation and contrast enhancement. IEEE Trans. Consum. Electron. 54 (August (3)), 1389–1397. Kim, J.Y., Kim, L.S., Hwang, S.H., 2001. An advanced contrast enhancement using partially overlapped sub-block histogram equalization. IEEE Trans. Circuits Syst. Video Technol. 11 (April (4)), 475–484.

Koh, T.S., Cheong, L.H., Hou, Z., Soh, Y.C., 2003. A physiologic model of capillarytissue exchange for dynamic contrast-enhanced imaging of tumor microcirculation. IEEE Trans. Biomed. Eng. 50 (February (2)), 159–167. Lai, C.C., Tsai, C.C., 2008. Backlight power reduction and image contrast enhancement using adaptive dimming for global backlight applications. IEEE Trans. Consum. Electron. 54 (May (2)), 669–674. Lamberti, F., Montrucchio, B., Sanna, A., 2006. CMBFHE: a novel contrast enhancement technique based on cascaded multistep binomial filtering histogram equalization. IEEE Trans. Consum. Electron. 52 (August (3)), 966–974. Oakley, J.P., Bu, H., 2007. Correction of simple contrast loss in color images. IEEE Trans. Image Process. 16 (February (2)), 511–522. Pei, S.C., Zeng, Y.C., Chang, C.H., 2004. Virtual restoration of ancient Chinese paintings using color contrast enhancement and lacuna texture synthesis. IEEE Trans. Image Process. 13 (March (3)), 416–429. Perner, P., 1999. An architecture for a CBR image segmentation system. Eng. Appl. Artif. Intel. 12, 749–759. Polesel, A., Ramponi, G., Mathews, V., 2000. Image enhancement via adaptive unsharp masking. IEEE Trans. Image Process. 9 (March (3)), 505–510. Schechner, Y.Y., Karpel, N., 2005. Recovery of underwater visibility and structure by polarization analysis. IEEE J. Ocean. Eng. 30 (July (3)), 570–587. Sherrier, R., Johnson, G., 1987. Regionally adaptive histogram equalization of the chest. IEEE Trans. Med. Imaging MI-6 (January (1)), 1–7. Shin, F.Y., Moh, J., Bourne, H., 1992. A neural architecture applied to the enhancement of noisy binary image. Eng. Appl. Artif. Intel. 5 (May (3)), 215–222. Sim, K., Tso, C., Tan, Y., 2007. Recursive sub-image histogram equalization applied to gray scale images. Pattern Recogn. Lett. 28 (July (10)), 1209–1221. Sun, X., Rosin, P.L., Martin, R.R., Langbein, F.C., 2009. Bas-relief generation using adaptive histogram equalization. IEEE Trans. Vis. Comput. Graph. 15 (July/ August (4)), 642–653. Tang, J., Kim, J., Peli, E., 2004. Image enhancement in the JPEG domain for people with vision impairment. IEEE Trans. Biomed. Eng. 51 (November (11)), 2013–2023. Wang, Y., Chen, Q., Zhang, B., 1999. Image enhancement based on equal area dualistic sub-image histogram equalization method. IEEE Trans. Consum. Electron. 45 (February (1)), 1186–1194. Xie, X., Lam, K.M., 2005. Face recognition under varying illumination based on a 2D face shape model. Pattern Recognition 38 (July (2)), 221–230. Yang, S.C., Wang, C.M., Hsu, H.H., Chung, P.C., Hsu, G.C., Juan, C.J., Lo, C.S., 2009. Contrast enhancement and tissues classification of breast MRI using Kalman filter-based linear mixing method. Comput. Med. Imaging Graph. 33 (January (3)), 187–196. Zhang, C.J., Wang, X.D., Duanmu, C.D., February 2010. Adaptive typhoon cloud image enhancement using genetic algorithm and non-linear gain operation in undecimated wavelet domain. Eng. Appl. Artif. Intel. 23, 61–73.