Contour Detection Method Based On Human Visual Perception

Contour detection refers to a procedure in digital image processing, in which both the influence of the textures from background and objectives and image noise are ignored, while the target contours are extracted via some technologies and methods. Therefore, contour detection is the foundation of target detection, shape analysis, target identification and target tracking etc..Human visual system can efficiently distinguish object contours from the entire scene, quickly fill the texture image and find the background noise in the image. Therefore, it is important to learn the mechanism of the human visual system, which will helps to improve the theories and applications of computer visioa In this dissertation, the research fruits of human visual system are employed to research on the new methods and models of contour detection. The main work includes:(1) After the definition, characteristic and progress of contour detection was described, the basic knowledge on the human visual perception system, especially its’ application in computer vision, are analyzed. Then, current contour detection methods based on human visual perception are summarized, and the pros and cons and the characteristic of edge detection based methods and visual perception based methods on contour detection are respectively discussed.(2) A contour detection method based on Window Edge Potential Function (WEPF) is proposed via the potential function in the classical electric energy. This method reduces the influence of image texture and image noise by the half-wave rectification of WEPF, and then employs two thresholds, and operators of dilation and erosion to obtain the final contours. Experimental results indicate the proposed method is superior to some edge detection based methods such as Canny, Prewitt and Sobel operators in terms of texture suppression and enhanced contours.(3)A contour detection method based on combined LoG filters (COLoG) is suggested. In this method, several LoG filters with different parameters are used to filter the original image, whose results not only are used to reduce the influence of texture and background, but also are introduced to enhance the contours of object via different weights. Experimental results show that compared with the traditional method based on Gabor function, the proposed method is able to suppress the texture effectively and ensure the integrity of object contours.