A natural image matting method based on visual saliency information
Sun We · Journal of Beijing Jiaotong University · 2015
To solve the problem of insufficient sampling in matting algorithm,we propose a color sampling method guided by visual saliency information and matting method based on sample pair confidence(SIGM).This method introduces visual saliency model into the sampling process and collect color salient points.We calculate the confidence of each sample pair using linear hypothesis,color space distance and local saliency as the basis for evaluation.Then we estimate the alpha value with high confidence sample pairs.Finally we use the matting Laplace matrix as the smooth item of the cost function to optimize the estimation alpha value and get the final matte.The experimental results show the effectiveness and accuracy of the proposed method.