Color space selection for color image segmentation by spectral clustering

Abstract

In this paper, we propose to segment color images by pixel clustering in a selected color space. This color space is selected among a set of classical color spaces according to a specific criterion based on a spectral clustering analysis. The clusters are determined by analyzing the 3D-histogram of the image coded in the selected color space thanks to a spectral clustering method. This scheme can identify clusters with complex shapes in the color space without any other post/pre-processing stages.