Combining Top-Down and Ncut Methods for Figure-Ground Segmentation

To locate the object accurately in a scene for further vision processing, a novel approach for figure-ground segmentation is proposed, which combines the normalized-cut method (Ncut) and top-down method inspired by the trickle-up and trickle-down processing in primate visual pathways. Firstly, as the trickle-up stage, the Ncut method groups the pixels into multiple partitions based on the global criterion, which measures both the total dissimilarity between the different groups as well as the total similarity within the groups. The computation of trickle-down includes mainly a covering operator, which covers the result of the trickle-up with the fragments of specific class. As one important computation in the trickle-down stage, the optimal method base on back-propagation neural network is utilized to improve the performance of the model. The proposed approach is applied to several segmentation experiments of clustering conditions. The results demonstrate that the performance of the proposed approach overpasses those achieved by previous top-down or bottom-up schemes on figure-ground segmentation. In addition to its application in computer vision, the success of this approach suggests a plausibility method, which combines the forward and backward processes for solving the visual perceptual grouping problem.