2017 3rd International Conference on Social Science and Technology Education (ICSSTE 2017) ISBN: 978-1-60595-437-0 A Novel Histogram-corrected Quadratic Histogram Equalization Image Enhancement Method Yi Zhang 1, Haifeng Wang Abstract The gradation level causes which in view of the traditional histogram balanced method is merged, the image partial bright, shortcomings and so on detail loss, this article had proposed one ind of histogram revises two histogram balanced image intensification method. The algorithm the traditional histogram balanced histogram according to the gradation level size of mappings to the original map histogram in, finally carries on liely to the new mapping histogram the equalization treatment. The experiment indicated this article algorithm retains in the image detail, reduces the gradation level to merge for superiorly, the visual enhancement effect is most gentle that appears especially in enhancement low light intensity infrared imagery time these superiority especially obviously. Keywords: Histogram Equalization; Information Entropy; Details of Reservation; Grayscale Merger 1. INTRODUCTION Image enhancement is the basic technique of image preprocessing. In many image enhancement methods, histogram equalization method (HE) has become a common method for image enhancement because of its simple and effective [1]. Although the histogram equalization algorithm has the advantages of fast operation and obvious enhancement effect, the gray scale is merged, resulting in the gray level reduction and loss of the image, the gray fault phenomenon, and the loss of the valuable part of the image and the visual effects blunt shortcomings [2]. In view of this shortcoming, many scholars have proposed some improvement methods: such as Hu Tianhai, who by extracting the original image of the highfrequency components and histogram equalization image superposition of the details of the enhanced method [3]. Gu Jianxiong et al. Used the power function to adjust the histogram probability distribution (for the large probability of compression, the probability of small stretch) after the histogram equalization, which can reduce the gray level was swallowed to retain the details [4]. Chen Wenfei, who tries to eep the information entropy in the same conditions, the use of distance control parameters to prevent the gray level was swallowed [5]. Song Yanfeng and others proposed a double-platform histogram correction of the equalization method and the histogram is too small to tae a fixed value T1, too large to tae T2, so as to reduce the gray level merger [6]. Han Dianyuan proposed an improved histogram equalization method, in determining whether the two gray levels are merged, according to the two gray levels of the distance were given a certain step size increase in the weight coefficient, reducing the gray level of excessive merger 1 Information Center, Jiangsu University of Technology, Changzhou, Jiangsu, 213001China 742
and better retains the details of the image [7], although the above method in the prevention of gray-scale annexation has achieved good results, any unsatisfactory. 2. Analysis of Traditional Histogram Equilibrium Principle and Its Defects To accumulate histogram, we can get the cumulative probability density function of the image as the follows. c( r ) P( ) 0 1 L 1 (1) r i i 0 Where the P( r ) n / N represents the probability of the gray scale r, while also is the histogram and the N is the total number of digital image pixel gray. The corresponding transform function can be obtained from the cumulative probability density function as the follows. T ( r ) ( L 1) c( r ) (2) According to the transformation function, each pixel in the input image can be mapped to each pixel to obtain a new pixel value, and the output image w can be obtained, and the contrast of the output image can be improved. From the formula (2) can be derived from the original image of the adjacent two gray-scale transformation of the distance. d T r ) T( r ) ( L 1) n / N ( L 1) P( r ) (3) ( 1 Thus, adjacent grayscale spacing after histogram equalization and associated probability density values of pixels in the input image. If the value is smaller, the spacing between the two is small, image contrast improvement is not obvious, the adjacent grayscale will merge, grayscale of the image would be reduced and if the value is bigger, the distance of adjacent grayscale will increase, the contrast is obvious increase. 3. The Mapping Process Description The image enhanced by traditional histogram equalization is the gray level will be reduced and this is because the original image gray scale caused by adjacent grayscale distance less than one consolidation, merger diagram as shown in figure 1. 743
Figure 1. Grayscale mergers and inserted into the diagram. Image grayscale probability P is less than 1/L-1, gray when histogram equalization are merged to the grayscale histogram equalization, cause of original image grayscale for total t effectively, through effective grayscale histogram equalization to q, and q < t. To minimize the loss of traditional histogram equalization grayscale, rich image detail, increase the gray levels, will histogram equalization in grayscale from big to small order mapping to the original histogram, form a new histogram after mapping, thus a good number of effective grayscale image, eliminate the "bright" phenomenon because of the enhanced image. 4. The Procedures of the Proposed Algorithm The algorithm to reduce the merger of grayscale histogram equalization method, enrich enhanced image gray levels, protect the details, the specific steps are as follows: 1. The histogram of the gray scale of the original image I is counted as: H I () 2. I carries on the traditional histogram equalizing to the original map to strengthen liely, after the enhancement the image is J, and counts its various gradations level histogram: H J () 3. Taing the H J () and H I () refer to the grayscale one-to-one mapping from big to small order, form a new histogram. 4. The new histogram H ' I ( ) is histogram equalized and arranged at equal intervals on the gray scale interval, and the output is taen as the enhancement image. 5. Experimental Analysis The experimental results show that the experimental results are standard Lena image and low illumination infrared image (Fig. 2). The experimental results are as follows: (1) The experimental results show that the experimental results are as follows: Figure 3, Figure 4 shows as the follows. 744
(a) lena image (b) Infrared image Figure 2. Original testing images. (a) Histogram equalization (b)[4] (c)proposed (d) Histogram equalization (e)[4] (f) Proposed Figure 3. Lena Image enhancement results compared. 745
(a) Histogram equalization (b)[4] (c)proposed (d) Histogram equalization (e)[4] (f) Proposed Figure 4. Low illumination infrared image enhancement result comparison. From the human subjective to normal gray image enhancement (as shown in figure 3), the traditional histogram equalization to enhance the image has been enhanced (bright), image detail is lost. Literature [4] algorithm and the algorithm of visual basic about the same on the feeling, no obvious difference but on the processing of the infrared images of low illumination (as shown in figure 4), the traditional histogram equalization to enhance the "bright" weaness particularly evident, image detail loss is very serious, literature [4] algorithm is better than traditional histogram equalization, but there are also too bright, some parts such as big deer body parts, compared with the former two inds of algorithm in this paper, the algorithm of visual effects, best details have maintained good, too has been obviously improved. Objective to verify the algorithm in detail the effect of retention and decrease grayscale been gobbled up, with the average brightness, effective number of gray levels, such as information entropy index to compare the algorithm as the follows. (1) The average brightness. Y P( ) 0 255 255 0 (2) Effective number of grayscale. S (if h ( ) 0 then S=S+1 0 255) 255 (3) Information entropy. E p i log pi i 0 The original image and the enhanced Y value are close to that, the brightness of the image enhancement algorithm to eep the better, otherwise poor. The larger the number of effective gray level S, the less the gray level of the image is enhanced, the better the details are 746
maintained; the other is that the gray level is swallowed up, and the detail is lost. The larger the value of information entropy E, the better the image detail is retained, the less the gray level of the annexation; otherwise, the details are lost and the gray level is swallowed. Lena image and infrared image enhancement algorithm results are shown in Table 1, Table 2. Table 1. Lena Image Test Result. Y S E Lena 124.61 216 2.2419 Histogram equalization 129.78 174 2.2122 [4] 125.91 176 2.2144 proposed 112.08 190 2.2162 Table 2. Infrared Image Experiment Data. Y S E Lena 62.41 211 1.8407 Histogram equalization 132.27 95 1.7727 [4] 70.32 97 1.7740 proposed 47.55 132 1.7751 From Table 1, Table 2 average brightness values looed, in three algorithms, although this article algorithm average brightness is smallest, but the visual effect quite is gentle. In prevented the effective gradation level in the performance which annexes, this article algorithm effective gradation progression quantity is biggest, lie in table 1, this article algorithm is 190, the traditional histogram balanced algorithm and the literature[4] algorithms are about 175, are many approximately 15 gradation levels; This merit displays in the infrared imagery prominently, lie Table in 2 low light intensity infrared imageries, this article algorithm effective gradation progression quantity are more than the first 2 algorithms approximately 40. Table 1, Table 2 information entropy in 3 algorithms all for biggest, explained this article algorithm strengthens the image detail retains well, simultaneously also indirectly explained this article algorithm in the image intensification process prevented the gradation level the performance which annexes is been most superior. In this paper, the traditional histogram equalization shows that the algorithm is the best to prevent the number of effective gray levels from being swallowed. From the information entropy size, in this paper, the algorithm is the smallest, the lowest brightness, but the softness of the visual effect, the above advantages in the enhanced low (Figure) [4] algorithm, 747
including traditional histogram equalization, indicating that the algorithm in the image detail to retain the best; from the image brightness point of view. Illumination when the infrared image is particularly evident. 6. Conclusion In view of the traditional histogram equalization method be merged, image gray levels caused by partial light, shortcomings and so on details missing, this paper proposes a histogram correction of the secondary image enhancement method of histogram equalization and the algorithm the histogram of the traditional histogram equalization, according to the size of grayscale one-to-one mapping to the original image histogram, finally, the new map histogram equalization processing. Through experimental comparison with other algorithms, data show that this algorithm eep, reduce the grayscale image details are combined for optimal, enhanced visual effect is the most gentle, especially when dealing with low illumination infrared images, the advantage is especially obvious. Acnowledgement This wor is supported by Changzhou Sci & Tech Program (No.CE20165049). References [1] Singh, Kuldeep, and Rajiv Kapoor. "Image enhancement using exposure based sub image histogram equalization." Pattern Recognition Letters 36 (2014): 10-14. [2] Lim, Sheng Hoong, et al. "A new histogram equalization method for digital image enhancement and brightness preservation." Signal, Image and Video Processing 9.3 (2015): 675-689. [3] Pandey, Kanchan, and Sapna Singh. "A comparative study of histogram equalization techniques for image contrast enhancement." IJESAT [International Journal of Engineering Science & Advanced Technology] 4 (2014). [4] Kaur, Amandeep, and Chandan Singh. "Contrast enhancement for cephalometric images using wavelet-based modified adaptive histogram equalization." Applied Soft Computing 51 (2017): 180-191. [5] Yadav, Garima, Saurabh Maheshwari, and Anjali Agarwal. "Foggy image enhancement using contrast limited adaptive histogram equalization of digitally filtered image: Performance improvement." Advances in Computing, Communications and Informatics (ICACCI, 2014 International Conference on. IEEE, 2014. [6] Kong, Nicholas, Sia Pi, Haidi Ibrahim, and Seng Chun Hoo. "A Literature Review on Histogram Equalization and Its Variations for Digital Image Enhancement." International Journal of Innovation, Management and Technology 4.4 (2013): 386. [7] Wang, Haoxiang, and Jingbin Wang. "An effective image representation method using ernel classification." Tools with Artificial Intelligence (ICTAI), 2014 IEEE 26th International Conference on. IEEE, 2014. 748
[8] Tarulatha, Balvant, Namrata Shroff, and M. B. Chaudhary. "VIBGYOR indexing technique for image mining." Data Mining and Advanced Computing (SAPIENCE), International Conference on. IEEE, 2016. [9] Kaur, Prabhjeet, and Kamaljit Kaur. "Review of Different Existing Image Mining Techniques." International Journal of Advanced Research in Computer Science and Software Engineering 4.6 (2014). [10] Dey, Nilanjan, et al. "Image mining framewor and techniques: a review." International Journal of Image Mining 1.1 (2015): 45-64. [11] Zhang, Shanwen, Harry Wang, and Wenzhun Huang. "Two-stage Plant Species Recognition by Combining Local K-NN and Weighted Sparse Representation." arxiv preprint arxiv:1702.04854 (2017). [12] Perer, Adam, Fei Wang, and Jianying Hu. "Mining and exploring care pathways from electronic medical records with visual analytics." Journal of Biomedical Informatics 56 (2015): 369-378. [13] Karimi, Sahar, and Fang Wang. "Online review helpfulness: Impact of reviewer profile image." Decision Support Systems (2017). [14] Bi, Chujian, Haoxiang Wang, and Rui Bao. "SAR image change detection using regularized dictionary learning and fuzzy clustering." Cloud Computing and Intelligence Systems (CCIS), 2014 IEEE 3rd International Conference on. IEEE, 2014. [15] Peng, Zhanglin, et al. "Deep boosting: Layered feature mining for general image classification." Multimedia and Expo (ICME), 2014 IEEE International Conference on. IEEE, 2014. 749