A novel image matting method using sparse manual clicks

Traditional image matting methods often have strict requirements on user input. This paper proposes a new matting method based on spectral clustering, which generates a well matte using a sparse user input. Firstly, connected components are obtained using spectral clustering, which actually utilizes a linear transformation of the smallest eigenvectors of the matting Laplacian matrix. An accurate trimap is obtained via user input and threshold segmentation. Secondly, sample sets are gathered by two-level hierarchical clustering and Fast Approximate Nearest Neighbors algorithm and unknown pixels are evaluated by the samples. Finally, an optimal matte is obtained by constructing an energy function with local smoothness constraint. Experiments show that the proposed method outperforms most of the state-of-the-art methods with a sparse user input and our method has fewer requirements to get a robust matte.