An Improved Fragment-Based Approach to Object Segmentation

The traditional segmentation method, that is, image-based segmentation method primarily use the continuity of grey-level, texture, and bounding contours. Although the method generates impressive results, however, it still often fails to capture meaningful and sometimes crucial parts especially when the ground is complicated and the shape of objects are variable. In this paper we utilize the current class-based segmentation method, which is guided by trained representation of figure-ground blocks of images within the same image class. Based on the class-based segmentation-CSF-SEG (Class-specific Fragment based Segmentation), we present a novel approach to extract fragments. The experimental results indicate the improved fragment-based segmentation approach works well for most images, and achieve more effective and robust segmentation than the current class-based segmentation.