This paper describes an efficient image matting method by combining color and depth information. First, the depth image is segmented by variational level set. Then morphological operators, dilation and erosion, are used to form trimap of ROI (Region of Interest). Finally, with preprocessed depth image, color image and trimap as inputs, an RGB-D Bayesian matting method is proposed to estimate the alpha matte. Experiments show our proposed RGB-D matting algorithm can significantly improve the quality of matting result with an automatically generated trimap.