Roughness Approach to Color Image Segmentation through Smoothing Local Difference

Aiming at the problems of histogram-based thresholding, rough set theory is applied to construct the roughness measure for segmenting color image. However, the extant roughness measure is a qualitative description of neighborhood similarity and tends to over focus on the trivial homogeneity. An improved roughness measure is proposed in this paper. The novel roughness is computed from smoothed local differences and quantified homogeneity, thus can form the accurate representation of homogeneous regions. The experimental results indicate that the segmentation based on improved roughness has good performances on most testing images.