Adaptive Centroid Placement Based SNIC for Superpixel Segmentation
Janith Bandara Senanayaka, Dilshan Morawaliyadda, Shehan Tharuka Senarath, Roshan Indika Godaliyadda, M. P. B. Ekanayake · 2020
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.