Spatial-color pixel classification by spectral clustering for color image segmentation

This work lies within the scope of color image segmentation by spatial-color pixel classification. Classes of pixels are difficult to be identified when the color distributions of the different objects highly overlap in the color space and when the color points give rise to non-convex clusters. We propose to apply spectral classification to regroup the pixels which represent the same regions, into classes. Spectral clustering achieves a spectral decomposition of a similarity matrix in order to construct an eigenspace in which the clusters are expected to be well separated. The originality of this paper lies in the similarity matrix which is derived from a spatial-color compactness function. This function takes into account both the distribution of colors in the color space and the spatial location of colors in the image plane. Experimental tests on synthetic and real images show that the spectral clustering technique associated with this spatial-color compactness function leads to promising results for segmentation purposes.