Research On Contour Detection Algorithm Based On Human Visual Information Processing Mechanism

Most information in an image is perceived from the edge of the image. When observing an image, what people firstly realize is the objects shapes in the image, i.e., the edge information in the image. Edge, the basic characteristic of an image, contains a variety of useful information in the image. It is very significant for edge detection to be widely used in such fields as image segmentation and pattern recognition, etc.. Since it tends to produce noise during the process of extracting image feature, it is necessary to remove noise and develop good edge detection technologies to ensure the accuracy of extracting image edge. Different from edge, contour is the subject outline feature extracted from the image, but not the edge in the background. Although the traditional edge detection methods have such advantages as fast processing speed and easy implementation, they do not take the context, the middle or even higher level information into account, and they cannot distinguish the difference between the object contour and the background edge. Therefore, it is very difficult to effectively extract object contour in an image. However, human vision system has the functions such as extracting object contour, distinguishing color and perceiving shape and orientation, which play a very important role in contour detection, In this paper, two kinds of contour detection methods are studied, which are based on the visual information processing mechanism. Some experiments were performed to analyze the performance of the two methods.Firstly, a contour detection method (CRFM model) is proposed based on the simulation of color opponent receptive field. In order to get the responses of retinal ganglion cells and LGN ones, method is adopted to simulate the properties of their single opponent receptive field in terms of the visual information processing. A novel method is presented to simulate their characteristics for contour detection based on the properties of double opponent receptive field in the primary visual cortex. Compared with other contour detection methods based on visual mechanism, the presented model can also get similar good contour detection results with higher efficiency, which makes it more practical in real world applications.Secondly, a contour detection method (CCM model) is presented based on color opponent receptive field and integration field. Based on the CORF model, The CCM model combines the theory of color opponent with the properties of integration field. Color opponent information is extracted, and then simple cell response is simulated by using the properties of integration field. The method can get the image contour with characteristics such as completer and smoother. In addition, the method can effectively extract objects contour, which is verified with a number of experimental results.In this paper, two kinds of contour detection models (CRFM and CCM), which are consistent with human visual mechanism, are established. They can not only extract target contour in an image more effectively, but also simulate human visual characteristics better. The work in this paper can help to further understand the visual principle, and plays important role in the engineering applications of image processing, analysis and computer vision.