Linear spectral clustering with mean shift filtering for superpixel segmentation

In this paper, we propose an improved algorithm for superpixel segmentation using linear spectral clustering with mean shift filtering. The proposed method adopts mean shift as pre-processing for superpixel segmentation. The proposed method uses this pre-processed image for spectral clustering, which produces more concise and uniform superpixels. Experimental results show that it gives superpixels with relatively compact and uniform segments. The proposed method reduces irregular and scratchy boundaries of segments compared with the existing superpixel segmentation methods.