Abstract: Color image segmentation is very essential and critical to modern image processing and analysis. According to the rough set theory, the lower, upper approximations and quantitative roughness representation of the color distributions are constructed based on the homogeneity of the pixels' neighborhood. Furthermore, a new segmentation approach --- QR measure is designed, in which an adaptive thresholding strategy is proposed to select peaks of indexes and merge regions. The experimental results demonstrate that the proposed approach outperforms several methods based on various kinds of statistics.