Adaptive centroid placement based snic for superpixel segmentation

Abstract

The proposed image segmentation algorithm identifies information-rich versus low information regions based on surface entropy. Thereafter on the information-rich regions, the mean shift algorithm is applied to generate possible centroid initialization points. This enables the Simple Non-Iterative Clustering (SNIC) algorithm when initialized through the proposed mechanism to provide more concentrated segmentation in those information-rich regions and sparse segmentation in the low information regions.