Edge Detection Algorithm Based on the Top-hat Operator Ying-Li WANG a, Shan-Shan MU b

Transcription

1 2016 Joint International Conference on Artificial Intelligence and Computer Engineering (AICE 2016) and International Conference on Network and Communication Security (NCS 2016) ISN: Edge Detection Algorithm ased on the Top-hat Operator Ying-Li WANG a, Shan-Shan MU Harin University of Science and Technology, China a @qq.com, @qq.com Keywords: Mathematical Morphology, Structural Elements, Top-hat Operator, Edge Detection. Astract. This paper proposes an image edge detection method ased on multi-directional, multi-scale Top-hat operators, and applies the method to the edge detection of OSAHS (Ostructive Sleep Apnea Hypopnea Syndrome) early pathological images. Firstly, construct multi-directional, multi-scale Top-hat operators, and they are used to detect the edge of image. Then the ideal image edge is otained y comining the edges of the image detected y each operator according to a certain weight, so that we can calculate the actual area of the oral cavity accurately, and then achieve electronic medical diagnosis. The simulation results show that the operator proposed in this paper can filter out the noise etter, preserve image detail more completely, so that the edge information of the image is more accurate and complete. Compared with conventional edge operator, it is more effective for image edge detection. Introduction Edge detection plays a vital role in image processing. It directly affects the results and accuracy of the image susequent processing. In recent years, mathematical morphology widely applied in image processing, covering the edge detection, texture analysis, image restoration and reconstruction and other fields. The asic idea of mathematical morphology is to use structural elements with certain shapes to detect an image. Structural elements are similar to the proe with orientation, size, color and other information. y moving to mark the locations which fit into the structural elements in the image; extract the structural characteristics of the image; achieve the purpose of analyzing and recognizing the image [1-3]. Traditional edge detection operators can e targeted to detect the image edge, ut have some limitations. In order to improve the prolems of traditional edge operators, in this paper, we applied the multi-directional, multi-scale Top-hat operator ased on mathematical morphology to detect the image edge. Compared with traditional edge operators, this method can filter out the noise etter, so that the edge information of the image is more accurate and more complete. Algorithm Principle and Implementation The asic Algorithm of Gray Morphology Gray dilation and gray erosion [4] are two asic algorithms of gray morphology. Let x y the input image, x y A, is, is the structural element. If D is the set of real integer, elongs to, A and are the pixel gray value functions corresponding coordinates, then gray dilation is defined as: ' ' ' ' ' ' ( A )( x, min A( x x, y y ) ( x, y ) ( x x )( y y ) D,( x, y ) Correspondingly, gray erosion is defined as: ' ' A D ' ' ' ' ' ' ( A )( x, max A( x x, y y ) ( x, y ) ( x x )( y y ) D,( x, y ) ' ' A D Among them, x and y are the domains of input image and structural element. Opening operation and closing operation are the different cominations of dilation and erosion. (1) (2)

2 Opening operation is firstly erosion and then dilation, denoted. Correspondingly, the closing operation is firstly dilation and then erosion, denoted. They are defined as: A ( A) A ( A ) Top-hat Operator Top-hat operator divided into opening Top-hat operators and the closing Top-hat operator. They are defined as: OTH A, ( x, A A (5) CTH A, ( x, A A (6) Top-hat operator has characteristics of high-pass filter. The high-frequency part is the edge of the image. Edge detection is to detect the high-frequency component of the image[5]. Therefore Top-hat operator has more advantages in the process of edge detection, It can effectively identify the target in various ackgrounds; extract a reasonale and accurate edge [6]. Select the Structure Elements Line, rectangle, octagon, disk, etc are the asic structural elements in the mathematical morphology. According to the characteristics of the input image select structural elements in the image processing. The input image is OSAHS early pathologic images in this paper. The images otained from clinical medicine have various shapes and contain noise. We select multi-directional, multi-scale structural elements for edge detection, in order to filter out noise etter; extract edge information of different types more accurately. Use the large-scale operators to detect overall changes; small-scale operators to detect slight changes of the image edge [7, 8]. In this paper, we take the edge detection of oral images for example; apply disk and line structural elements in different directions and scales. Structure elements in different scales can e otained y using morphological dilation operation. Namely, large-scale structural elements can e otained y using morphological dilation operation of small-scale structural elements. Assume that m is a finite structure element, then multi-scale structural elements can e defined as: m m Among them, m is the scale parameter, takes a positive integer. Suppose 1(a) is the minimum 3 3 disk structure element, then 1() and 1(c) are 5 5 and 7 7 disk structural elements, and so on, 1(c) is the disk structure element. As shown in Figure (a) Structural Element 1 () Structural Element 2 Figure 1. Multi-scale Structural Elements (c) Structural Element 3 Using simple symmetrical disk structural elements to detect image edge would reduce the sensitivity to different directions of the edge, which would result in the loss of the direction details of the image edge. In order to improve the sensitivity of the edge information in different directions; (3) (4) (7)

3 detect complete and accurate edge [9], we construct the structural element in four directions in this paper. The structural element is shown in Figure 2. Improve the Top-hat Operator Figure 2. Multi-directional Structural Element 1. The traditional morphological Top-hat operators use the same structural elements to complete the opening and closing operations, they have no effects on the suppression of noise, and they will lead to the lack of edge information. It is the most important to select the structural elements reasonaly in the edge detection processing. Small-scale structure elements can extract the edge details accurately; large-scale structure elements can filter out the noise effectively; multi-directional structure elements can improve the sensitivity of the edge information in different directions [10]. In order to fuse the advantages of the structural elements in different directions and scales, we proposed an integrated Top-hat operator ased on structural elements in different directions and scales, OTH ( x, ( A ( A ) )( x, OTH ( x, ( A ( A ) )( x, Among them, i, i k l j k are small-scale structure elements, j, (8) (9) l are large-scale structure elements. Use each Top-hat operator to detect the edge of the image, then comine each edge with others which detected y operators according to a certain weight to get an ideal image edges. Experiment and Analysis (a) Original Image () The Result of Roerts Operator (c) The Result of Prewitt Operator (d) The Result of Log Operator (e) The Result of Canny Operator (f) The Result of this Paper s Operator Figure 3. Oral Image Processing Result. Figure 3 shows the results of edge detection y Roerts operator, Prewitt operator, Log operator, Canny operator and the Top-hat operator which proposed in this paper. 3(a) is the original image, 3() is the result of edge detection y Roerts operator, 3(c) is the result of edge detection y Prewitt operator. 3(d) is the result of edge detection y Log operator. 3(e) is the result of edge

4 detection y Canny operator. 3(f) is the result of edge detection y the Top-hat operator which proposed in this paper. As we can see from the figure, Roerts operator and Prewitt operator extracted the edge inaccurately; didn t detect the complete edge information; Log operator and Canny operator are sensitive to the noise, they extracted a lot of pseudo-edge information; The algorithm which proposed in this paper can detect the edge information more accurately than other operators, and have a good Edge closure. Figure 4 shows the results of edge detection y Roerts operator, Prewitt operator, Log operator, Canny operator and the Top-hat operator which proposed in this paper after adding salt and pepper noise to the original image. 4(a) is the original image, 4() is the result of edge detection y Roerts operator, 4(c) is the result of edge detection y Prewitt operator. 4(d) is the result of edge detection y Log operator. 4(e) is the result of edge detection y Canny operator. 4(f) is the result of edge detection y the Top-hat operator which proposed in this paper. As we can see from the figure, Roerts operator and Prewitt operator seemed that they were no use filtering out the noise, they detected the portion of the edge; Log operator could suppress the noise oviously, at the same time, it suppressed the original edge information, didn t detect the complete edge information. Canny operator could also suppress the noise oviously, at the same time, it smoothed out some of the edge information, leading to the edge of the image was not clear. The algorithm which proposed in this paper can filter out the noise etter and detect the edge information more accurately than other operators, and have a good Edge closure. (a) Original Image () The Result of Roerts Operator (c) The Result of Prewittoperator (d) The Result of Log Operator (e) The Result of Canny Operator (f) The Result of this Paper s Operator Figure 4. Oral Image Added Noise Processing Result. Tale 1 presents the performance comparison of these algorithms. As we can see from the Tale, use this paper s operator to detect the edge, the closure edge is etter than the edges which detected y other operators. Tale 1. Performance Comparison of Several Operators. Roerts Prewitt Log Canny This paper s operator operator operator operator operator Edge closure 21.5% 38.3% 67.8% 85.5% 98.7% Conclusions This paper proposed a Top-hat operator ased on the multi-directional, multi-scale structural elements. After oral medical image preprocessing, the ideal image edge is otained y comining the edges of the image detected y each operator according to a certain weight, so that we can

5 calculate the actual area of the oral cavity accurately, and then achieve electronic medical diagnosis. Compared with conventional edge operator, the operator proposed in this paper can filter out the noise etter, preserve image detail more completely. It is very useful. Scientific Research Fundation of the Education Department of Heilongjiang Province, , China. References [1]Huang HaiLong, Wang Hong, Ji Li. Order morphology edge detection algorithm ased on partial fuzzy enhancement [J]. Chinese Journal of Scientific Instrument, 2012, 33(11): [2]ai Xiangzhi, Zhou Fugen, Xue indang. Noise-suppressed image enhancement using multiscale Top-hat selection transform through region extraction. Applied Optics, 2012, 51(3): [3]Wu YiQuan, Song Yu, Zhou HuaiChun. Edge detection of comustion flame images ased on anisotropic mathematical morphology [J]. Chinese Journal of Scientific Instrument, 2013, 34(8): [4]Gonzalez, R.C., R.E. Woods, and S.L. Eddins. Digital Image Processing Using MATLA. Gatesmark Pulishing, [5]Zhang LiGuo, Yang Jin, Li Jing, et al. Image characteristic extraction method ased on wavelet packet and mathematical morphology [J]. Chinese Journal of Scientific Instrument, 2010, 31(10): [6]Zeng Ming, Li JianXun. The small target detection in infrared image ased on adaptive morphological Top-hat filter [J]. Journal of Shanghai Jiaotong University, 2006, 40(1): [7]C. Dueottet, T. Fournel and C. arat. Seale-adaPtive deteetion and local characterization of edges ased on wavelet transform, Signal Proeessing, 2004, 84: [8]Felice Andrea Pellegrino, Wlater VanZella and Vincent Torre. Edge detection revisited, IEEE Trans. Systems, Man and Cyemetics-Part: Cyernetics, 2004, 34(3): [9]Zhao Yuqian, Gui Weihua, Chen Zhencheng. Edge Detection ased on Multi-Structure Elemenets Morphology. The 6th Proceedings of the World Congress on Intelligent Control and Automation (WCICA), Dalina, China, [10]ai Xiangzhi, Zhou Fugen, Xue indang. Fusion of infrared and visual images through region extraction y using multi scale center-surround Top-hat transform. Optics Express, 2011, 19(9):