Alpha estimation in perceptual color space

This paper presents an efficient algorithm for alpha estimation based on perceptual color space in natural image matting. A user specifies a trimap which partitions an image into definitely foreground, definitely background and unknown. For each pixel in the unknown region, the algorithm separates intensity and chroma information from RGB color after its foreground and background color components are estimated. Consequently, the alpha component is decomposed into intensity part and chroma part. The weighted mean of these two parts yields the final alpha value. Experimental results demonstrate the efficiency and advantage of the proposed algorithm over other approaches.