Fig. 1. Medical Image Enhancement. Input Image. Pre-Processing. Filter method. Post processing

Transcription

1 International Journals of Advanced Research in Computer Science and Software Engineering Research Article June 2017 A Comparative Analysis on Histogram Equalization Techniques for Medical Image Enhancement Himanshu Singh *, Assistant Prof.Vivek Singh Dept. of CSE/IT, ITM University Gwalior, Madhya Pradesh, India DOI: /ijarcsse/V7I6/0269 Abstract Image enhancement (IE) is such technique which makes the processing on image in order to increase its effectiveness for computer to process. Enhancement is, used for improvement of the visual effects and the clarity of image.medical image is used as the important information and basis for the clinical diagnosis, while its quality digressed because of the interferences caused by the human body structure, equipment s, and environmental factors. In this paper review, the enhancement of medical images using efficient algorithms based on HE techniques. Also evaluate the result of medical image on technique to histogram equalization (HE), adaptive histogram equalization (AHE), (BBHE) and contrast limited adaptive histogram equalization (CLAHE) to compare the pre-processing of image. Keywords Image Enhancement; Histogram Equalization; Adaptive histogram equalization; Contrast limited adaptive histogram equalization, PSNR; Entropy. I. INTRODUCTION In modern world due to the changing life style of human beings, everyday new types of diseases are emerging. It s an everyday challenge for doctors to effectively diagnose these diseases and provide remedy [1]. In today s medical diagnosis such a crucial part is played by Medical images. Medicinal pictures have a critical influence in today's restorative finding. Medicinal imaging innovations, for example, Computerized Tomography (CT), and X-ray imaging give clear and direct perspective of the neurotic ranges. They are fundamental apparatuses for distinguishing and diagnosing different maladies. In any case, because of the constraint of imaging equipment, got therapeutic pictures frequently introduce low determination or low difference. Medicinal picture upgrade intends to enhance restorative picture difference or accentuation certain elements. It is important to build the recognition rate of different ailment and it has been one of the key research zones of advanced picture preparing [2]. Several image enhancement methods can be used: One is the enhancement method based on histogram equalization (HE: HE is a common method of IE, but which is unable to be effective enough in practical applications [3]. Thus the researchers put forward a series of improved method, Pizer [4] proposed adaptive histogram equalization algorithm (AHE), which use a window sliding on the image to calculate the local gray level histogram, mapping the center pixel of the window. This algorithm makes full use of the field information, but needs to calculate the histogram distribution in each window, which leads to the low efficiency and noise sensitivity. Fig. 1. Medical Image Enhancement Input Image Pre-Processing Filter method Post processing Enhanced Image Fig. 2. Enhancement Model All Rights Reserved Page 364

2 Zuiderveld [5] proposed contrast limited adaptive histogram equalization (CLAHE) algorithm on the basic of AHE to avoid noise sensitivity. Many other enhancement methods are developed over the years such as brightness preserving bi-histogram equalization (BBHE), bi- gray level grouping (GLG). This algorithm has improved performance but did not remove the issues. In the below section II present the related research work of medical image enhancement under histogram techniques, section III present the different techniques of histogram equalization with some evaluate result. Where as in Section IV shows applications areas of image enhancement and performance metrics on which enhancement result computed. II. LITERATURE REVIEW Anita Thakur (2015) et al introduce that Human visual framework conciliates by a decent differentiation Images. Picture upgrade strategies are best answer for enhancing the visual appearance of pictures to a human watcher. It likewise saves the structure elements of the picture. Upgrade of the noisy picture information without losing any critical data is exceptionally testing. There are numerous instabilities included while capturing picture and the execution of picture improvement fluctuates with subject. It is entrenched that Fuzzy logic and fuzzy sets are great at taking care of numerous vulnerabilities. [6]. M.Shakeri (2016)- Histogram division and playing out a different leveling for each sub-histogram is one of the displayed arrangements. The partitioning technique and deciding the quantity of sub-histograms are the fundamental issues specifically influencing the yield picture quality. In this review, a technique is presented for automatic determination of the quantity of sub-histograms and thickness based histogram division prompting fitting yield with no requirement for parameter setting. Every principle pinnacle is in a different area. Image Contrast is expanded with no loss of picture details through deciding the quantity of sub-histograms in view of the quantity of fundamental pinnacles.[7] He Wen (2016) presented a picture improvement calculation which in light of wavelet domain homomorphic filtering and CLAHE. The picture is divided by DWT; the picture is disintegrated into low-frequency and high-frequency coefficients of first layer of wavelet space. At that point the low frequency coefficients are prepared by an enhanced homomorphic channel, and after that direct increased [8]. Se Eun Kim (2016) presents an entropy-based IE technique in the wavelet space. This strategy isused in the HIS shading space. The low-frequency coefficients in the wavelet area are altered by the worldwide histogram-based approach. The high-frequency coefficients are scaled by amplifying the entropy of the complexity characterized in the wavelet area [9]. Cheolkon Jung (2016)- In this work, it proposed a powerful difference upgrade technique in light of dual tree complex wavelet transform (DT-CWT) to work on an extensive variety of symbolism without IE. In the terms of improvement, it utilized the nonlinear reaction of the human eye to the luminance to outline a logarithmic capacity for worldwide shine advancement. Additionally, the nearby difference is upgraded by CLAHE in low-pass sub groups, which makes the structure of picture clearer [10]. Jing-Wein Wang (2016) to manage these troubles, an illumination compensation technique, versatile SVD in the twodimensional discrete Fourier domain (ASVDF) and a productive brilliance indicator for lighting detection, for face picture improvement are proposed in this method. The outcomes for the CMU-PIE, Color FERET, and FEI confront databases demonstrate that the technique extensively enhances the nature of face pictures, even sidelong lighting, accordingly enhancing the exactness of face recognition generously [11]. Qiong Song (2016) - They exhibited another way to deal with show HDR IR pictures with IE. In the first place, the local edge-preservingfilter (LEPF) is used to isolate the picture into a base layer and detail layer(s). After the separating technique, it utilized a versatile Gamma change to alter the gray dissemination of the base layer, and extend the detail layer in light of a human visual impact guideline. At that point, we recombine the detail layer and base layer to get the improve output [12]. G.N.Sagar et al. [2012], proposed IE technique with filtering techniques which shows the enhancement in the contrast of x-ray images which were manipulated because of the noise and blurring. They use different filtering techniques to remove noise such as median filter and to remove the high frequency details,mean filter is used. They used MSE and PSNR to measure the performance of the proposed technique. Not only removing noise but it also has the capability to increase X-ray image quality. But it has a lot of improvement to eliminate the noise in the X-ray images completely [13]. Sundaram et al. [2011], proposed the Histogram Modified CLAHE (HM CLAHE). It can manage the level of complexity improvement, which frequently give the resultant picture as a strong contrast and brings the area points of interest for understanding that is more important. It partners both histogram alterations as an optimization method and CLAHE[14]. [2013], presented a neuro-fuzzy inference system to produce the enhanced image. HE is used for contrast enhancement. [15]. Nercessian et al. [2013], proposed aie strategy with the preferences: the coordination of both luminance and difference masking phenomena, and a prompt method for altering overall brightness, and accomplishing dynamic range pressure IE coordinate multi-scale improvement structure. Test comes about showed the capacity of the proposed calculation to accomplish concurrent nearby and worldwide upgrades [16]. III. TECHNIQUES OF HISTOGRAM EQUALIZATION A. Histogram Equalization HE is the highest operated contrast enhancement technique due to the easiness comfort.he makes density scattering flatter and it expanses the gray level range to enhance the total contrast of the given image. Transformation of gray levels All Rights Reserved Page 365

3 of the given image to its improved image level cumulative distribution function (CDF) has been utilized by this technique. HE changes the mean brightness of image to middle of overall dynamic range, which is a drawback of HE because of intensity saturation and annoying artifacts occurs [17]. (a) Input Image (b) Histogram Image Fig. 3. Histogram Equalization B. Adaptive Histogram Equalization AHE is different from the normal HE method because HE gives only one histogram but AHE method generates several histograms corresponding to different area of the image and by using that it redistributes the intensity values of the image. By using ADE method can improve the detection of spiculation on dense mammographic backgrounds. Saeid et al. [18], proposed an AHE method for segmentation of blood vessels in color retinal images. There are two methods of retinal vessel segmentation and these are first derivative of Gaussian matched filter and the other one is simple Gaussian matched filter. AHE gives the improved results of contrast of image which enhance the quality of image of retinal vessels used in identification of diseases like high blood pressure and diabetes. This method gives the raise of about 2 percent in accuracy as compared to previous methods with an accuracy of Analysis of this approach shows that AHE method used in retinal vessel segmentation is based on threshold. Therefore, this approach is suitable for specific type of images. (a) Input Image (b) Adaptive Histogram Image Fig. 4. Adaptive Histogram Equalization C. BI- Histogram Equalization TheBrightness Preserving Bi-Histogram Equalization (BBHE) is a type of technique in which the image histogram is divided into two parts. This is the type of method in which, partitions intensity is given by the mean brightness value of input image, and this intensity is the average intensity of all pixels that combines to make the image given by input. After that the BBHE autonomously makes the sub-picture square with as indicated by their applicable histograms in the constraint that the correct set's samples ought to be mapped into the range from the base gray-level to the information mean and the examples in the last set ought to be mapped into the range from the mean of the greatest gray level. Therefore, the result of balanced sub-pictures is encompassed by each other around the info mean, which gives the consequence of safeguarding mean brightness. [19]. (a) Input Image (b) BBHE Image Fig. 5. Brightness Preserving Bi-Histogram Equalization (BBHE) D. BI- Gray Level Grouping (GLG) The basic principle involved in this technique is as follows. Firstly, we arrange the histogram components in groups with respect to a proper number of gray level bins according to their amplitudes for the reduction of the number of gray bins. The main objective of this technique is to get a uniform histogram for a low contrast color image. Conventional All Rights Reserved Page 366

4 histogram equalization results in under or over contrast image since it leaves too much empty space on the grayscale. The drawback of GLG is that it is not computationally efficient compared to fuzzy-based methods. The quantitative analysis represent that fuzzy-based methods are superior to GLG [19]. E. Brightness Preserving Dynamic Histogram Equalization (BPDHE) In the advancement of HE and DHE, proposed new method BPDHE. In DHE, the histogram of input image is partitioned into small parts which are called sub-histograms. This method is useful to providing brightness of an image and which gives the new range to intensities [20]. The realistic image is provided by its look. In this method, calculate the mean brightness between resultant image and input image which is shifted by BPDHE, which preserves the mean brightness. And it equalizes the average intensity of input and output images. Many different filters are used by BPDHE such as Gaussian filter,smoothing filter, etc. which makes the data smooth by suppressing the noise of image in order to get clear image [21]. F. Contrast-Limited Adaptive Histogram Equalization (CLAHE) CLAHE differs from ordinary AHE as it limits the contrast. This feature is also applicable to global HE, which gives rise to CLAHE. It is mostly used in enhancement of low contrast retinal image. In case of CLAHE, a transformation function derived from contrast limited procedure to each neighborhood pixel. [17]. (a) Input Image (b) CLAHE Image Fig. 6. Contrast-Limited Adaptive Histogram Equalization (CLAHE) G. Adaptive DWT (ADWT) baseddynamic Stochastic Resonance( DSR) A DWT is simply like other wavelet transform which uses wavelet coefficients. By using this technique the content image of high frequency is produced. The DWT further divides the input image to form sub bands. These bands are referred as High-High (HH), Low-High (LH),Low-Low (LL), and High-Low (HL). The image processing using DWT is done by introducing high-frequency sub band images to the low-resolution original images which gives the output as the improved image. The ADWT based DSR is a technique which is proposed to enhance the very dark images. This gives much improve performance in the enhancement of very dark images. Also the computational complexity is very low in this method. [22]. The difference between DSR and ADWT based DSR is that DSR uses external noise of an image and the ADWT based DSR utilizes inside noise to give better execution of an input picture. It produces the yield without ringing, curios, obstructing of the picture. This method of adding noise to the input picture is extremely helpful in the event of nonstraight frameworks. Because in SR component the flag can't have the capacity to achieve the limit an incentive by utilizing lower noise, so it requires noise which enables the flag to achieve the edge esteem. In this manner ADWT based DSR is reasonable for improvement of both the grayscale and colored picture. IV. APPLICATIONS IE is utilized to improve the image quality. IE are used in these applications namely, Aerial imaging, Satellite imaging, Medical imaging and in remote detecting. IE strategies utilized as a part of numerous regions, for example, crime scene investigation, Astrophotography and in Fingerprint coordinating, and so on. [19]. V. PERFORMANCE MEASURES The performance parameters are histogram, entropy, SNR and PSNR. A. Entropy The average information content known as entropy is used for measure of image quality. Larger the value of entropy, more the information content in the image. L 1 ENT = i=0 I l logl(l) (1) Where ENT (i) denotes entropy, I(l) is pdf of image having l intensity level and L denotes the number of gray levels. B. Peak signal-to-noise ratio (PSNR) For PSNR calculation first mean square error (MSE) is calculated as- N j =1 X i,j Y(i,J ) 2 MSE = (2) M N The root mean square error (RMSE) is calculated from root of MSE then PSNR as- M i=1 All Rights Reserved Page 367

5 max (Y i,j ) PSNR = 20. log 10 ( ) (3) RMSE Here, X (i, j) is input image having M by N pixels, Y(i, j) is enhanced image. Greater the PSNR better will be contrast of enhanced image. C. Signal to Noise Ratio (SNR): Consider r(x,y) be the original image and t(x,y) is enhanced image. The noise estimation in enhanced fundus image is analyzed by- SNR = 10. log 10 1 n x n y n x 1 n y r (x,y) 2 n x 1 n y r x,y t(x,y) 2 VI. COMPARISON OF TECHNIQUES [23] Sr. No Technique Concept Advantage Disadvantage 1 HE Uniform distribution of gray values over scale. Simple, effective and low complexity. Brightness of an image 2 BBHE Decomposition of image using mean value. 3 Gray Scale Grouping 4 AHE Formation of bin of grey values Preserve the brightness of an image. Applicable to a broad variety of images. AHE tends to over open up noise in moderately homogeneous districts of a picture. (4) Gives an artificial look to image. More Complex Complex and enhances high contrast area much more. Original Image HE AHE BBHE CLAHE PSNR= PSNR= PSNR= PSNR= PSNR= PSNR= PSNR= PSNR= PSNR= PSNR= PSNR= PSNR= PSNR= PSNR= PSNR= PSNR= All Rights Reserved Page 368

6 PSNR= PSNR= PSNR= PSNR= VII. CONCLUSION Medical image is used as the important information and basis for the clinical diagnosis. In order to enhance the diagnosis quality and accuracy, medical image enhancement is very necessary and also an important research direction of biomedical engineering. Most commonly used image enhancement method for haze-removing can be applied in medical images, in this paper review, the enhancement of medical images using efficient algorithms based on HE techniques. Also evaluate the result of medical image on technique HE, AHE, BBHE and CLAHE to compare the pre-processing of image. REFERENCES [1] Vidyasaraswathi H N, Dr. M.C.Hanumantharaju "Review Of Various Histogram Based MedicalImage Enhancement Techniques" ICARCSET '15, March 06-07, 2015, Unnao, India, ACM [2] Zhou Zhao, Yicong Zhou Comparative Study of Logarithmic ImageProcessing Models for Medical Image Enhancement International Conference on Systems, Man, and Cybernetics, IEEE2016. [3] Wang Rui, Wang Guoyu Medical X-ray Image Enhancement Method based on DarkChannel Prior ICBCB '17, January 06-08, 2017, ACM., [4] Pizer S. Adaptive histogram equalization and its variations [J].Computer Vision, Graphics and Image Processing, 1987, 39(3): [5] K. Zuiderveld. Contrast Limited Adaptive Histogram Equalization. In: P. Heckbert: Graphics Gems IV, Academic Press 1994 [6] Anita Thakur and Deepak Mishra, Fuzzy Contrast Mapping for Image Enhancement, nd International Conference on Signal Processing and Integrated Networks (SPIN), pp: [7] M.Shakeri, M.H.Dezfoulian, H.Khotanlou, A.H.Barati, Y.Masoumi, Image contrast enhancement using fuzzy clustering with adaptive cluster parameter and sub-histogram equalization, / 2016 Elsevier. [8] He Wen, Wu Qi, Li Shuang, Medical X-ray Image Enhancement Based on Wavelet Domain Homomorphic Filtering and CLAHE, 2016 International Conference on Robots & Intelligent System IEEE, pp [9] Se EunKim,JongJuJeon,IlKyuEom, Image contrast enhancement using entropy scaling in wavelet domain, /& 2016 Published by Elsevier B.V. [10] Cheolkon Jung, Qi Yang, Tingting Sun, Qingtao Fu, Hyoseob Song, Low Light Image Enhancement with Dual-Tree Complex Wavelet Transform, J. Vis. Commun. Image R. (2016) [11] Jing-Wein Wang, Ngoc Tuyen Le, Jiann-Shu Lee,Chou-Chen Wang, Color Face Image Enhancement Using Adaptive Singular Value Decomposition in Fourier Domain for Face Recognition, Pattern Recognition 2016, [12] Qiong Songa, Yuehuan Wanga,b,_, Kun Baia, High dynamic range infrared images detail enhancement based on local edge preserving filter, Infrared Physics & Technology (2016). [13] Jha, Rajib Kumar, Rajlaxmi Chouhan, Prabir Kumar Biswas, and Kiyoharu Aizawa. "Internal noise-induced contrast enhancement of dark images." In Image Processing (ICIP), th IEEE International Conference on, pp IEEE, [14] Chauhan, Ritu, and Sarita Singh Bhadoria. "An improved image contrast enhancement based on histogram equalization and brightness preserving weight clustering histogram equalization." Communication Systems and Network Technologies (CSNT), 2011 International Conference on. IEEE, [15] Maragatham, G., and S. Md Mansoor Roomi. "An automatic contrast enhancement method based on stochastic resonance." In 2013 Fourth International Conference on Computing, Communications and Networking Technologies (ICCCNT), pp IEEE, [16] Jaspreet Singh Rajal An Approach for Image Enhancement Using Fuzzy Inference System for Noisy Image Journal of Engineering, Computers & Applied Sciences (JEC&AS) ISSN No: , 2013, Volume2,No.5,pp [17] Sharad Kumar Yadav1, Shailesh Kumar2, Basant Kumar3, Rajiv Gupta4 Comparative Analysis of Fundus Image Enhancement in Detection of Diabetic Retinopathy. [18] SaeidFazli, SevinSamadi, ParisaNadirkhanlou, A noveletinal vessel segmentation based on local adaptive histogram equalization, th Iranian Conference on Machine Vision and Image Processing. [19] Sargun and Shashi B. Rana A Review of Medical Image Enhancement Techniques for Image Processing International Journal of Current Engineering and Technology, Vol.5, All Rights Reserved Page 369

7 [20] Kong.N.S.P, Ibrahim.H, Color Image Enhancement using Brightness Preserving Dynamic Histogram Equalization, IEEE Transaction on Communication, Networking and Broadcasting, Page: , Publication year: [21] Kuo-Liang Chung, Yu-Ren Lai, Chyou-Hwa Chen, Wei-Jen Yang, and Guei-Yin Lin, Local Brightness Preservation for Dynamic Histogram Equalization, [22] P.Suganya, S.Gayathri and N.Mohanapriya, Survey on Image Enhancement Techniques, International Journal of Computer Applications Technology and Research Volume 2 Issue 5, , [23] Rashmi Choudhary, Sushopti Gawade Survey on Image Contrast Enhancement Techniques International Journal of Innovative Studies in Sciences and Engineering Technology(IJISSET)Volume: 2 Issue: 3 March All Rights Reserved Page 370