A Foreground Segmentation Algorithm Based On Active Contour Model And Binocular Disparity

Image segmentation is an essential step for objects detection, recognition and tracing, which is one of hot points in computer vision. In the past 20 years, the theoretical and experimental research mostly devoted to the monocular image segmentation. Moreover, active contour model has been widely applied to objects segmentation in monocular images due to its merits such as easy modeling and high efficiency in mathematical solution. On the other hand, stereoscopic images have been utilized successfully in many fields such as industrial detection, automatic navigation and medical auxiliary. Accordingly, more and more scholars paid attention to research on stereoscopic image segmentation. This paper presents an automatic segmentation for color stereoscopic image with complex background, which is based on the active contour model and utilizes the adaptive initial contour extraction. It may provide valid theoretical support for the applications based on stereoscopic image, for example, automatic detection and automatic recognition, etc.So far, many effective active contour models have been proposed. The local Chan-Vese(LCV) model is widely used because of it can deal with automatic segmentation of weak edge and accommodate itself to the changes of the object’s Geometric topology. Even that model, however, involves some drawbacks. For instance, it is sensitive to initial contour position and the image is divided into different regions without reference standards. The most important thing is that LCV model is only applicable to monocular grayscale image segmentation, not suitable for color stereo image segmentation. To solve the problem above, an adaptive segmentation based on active contour model and binocular disparity in the stereoscopic images is presented. The method consists of two parts: selecting appropriate initial position and the curve evolution. Firstly, the binocular disparity is utilized to identify the target regions in the stereoscopic images and then an initial contour is formed adaptively and automatically by the identified target regions. Then, curve evolution which is based on the level set method and local image. Notice that the local image not only preserves the foreground’ details but also removes the most of background information.Compared with conventional active contour models, the proposed method can describe the initial curve adaptively by virtue of disparity values so as to decrease greatly the impact of initial contour position on final segmentation results. Furthermore, the proposed model reduces the numbers of interaction of contour curve evolution so as to improve the efficiency of image segmentation. The experiment results show that the proposed method can effectively segment stereoscopic images.