Multi-Iterative Superpixel Segmentation based on Local Brightness and Darkness Information

Super-pixel segmentation algorithms are widely used in the preprocessing steps for computer vision applications. A crucial aspect of Super-pixel segmentation is preserving structure boundaries. However, in many images with complex structures, the contrast between foreground and background are very low, which becomes a challenge for Super-pixel segmentation. In this paper, we introduce a new method to evaluate local brightness and darkness for gray images. Then, we use local brightness and darkness information to enhance the weak structure boundary for Super-pixel segmentation. Furthermore, we also introduce a Super-pixel merging method for SLIC to eliminate numbers of Super-pixel blocks, especially nearby the boundaries between different objects. The experimental results show the proposed algorithm makes Super-pixels adhere to object boundaries better and improve the over-segmentation.