||Linear Spectral Clustering with Contrast-limited Adaptive Histogram Equalization for Superpixel Segmentation
||Changhoon Yim; Jiyeon Baek;Yejin Kim;Byungjin Chung
|| Image segmentation; Superpixel; Clustering; Mean shift; Histogram equalization
||This paper proposes an improved superpixel segmentation method using linear spectral clustering (LSC) with contrast-limited adaptive histogram equalization as a pre-processing step for clustering. It restores the global structure of an image so that the boundary coherence of the clustering result is improved. The algorithm also performs mean shift filtering after contrast-limited adaptive histogram equalization. This removes noise caused by histogram equalization. In LSC, a simple modification of the initialization method is proposed for k-means clustering. Experimental results demonstrate that the proposed method generates superpixels with higher boundary coherence compared to previous superpixel segmentation methods.