Trail-Dependent Intelligent Scissors Based on Multi-Scale Image Segmentation
Yi‐Ping Hung, Yu-Pao Tsai · 2002
Image segmentation is a very important topic in computer vision. However, due to the large variation of image content, fully automatic image segmentation for general applications is still an open problem. Therefore, our goal is to develop an interactive image segmentation tool that can accurately extract the desired object boundaries with minimal human efforts. In this paper, we propose a new trail-dependent intelligent scissors, which let the user interactively extract desired object boundaries based on multi-scale image segmentation. By utilizing the information contained in the trail of the cursor’s motion, which somewhat implies the intention of the human operator, our intelligent scissors can allow the user to extract a desired object boundary with less mouse-clicking, and hence is more user-friendly. This is the major advantage of our new intelligent scissors. Another advantage is that our intelligent scissors permits the user to trace the object boundary with less tension by utilizing the coarse-to-fine region boundaries provided by multi-scale image segmentation. Our experiments have demonstrated that the new interactive segmentation tool requires less human efforts than the previously available tools.