Grouping optimization algorithm for natural image matting
Shuang Liu, Fujian Feng, Hongshan Gou, Zhulian Zhou, Man Tan, Lin Wang · 2021 IEEE International Conference on Computer Science, Electronic Information Engineering and Intelligent Control Technology (CEI) · 2021
Image matting is an important technology for image synthesis and video editing. The existing image matting method based on pixel pair optimization can significantly improve the accuracy of image matting, but with the increase of image resolution, the optimization difficulty increases. This paper proposes a group optimization strategy, which transforms the large-scale combinatorial optimization matting problem into multiple small-scale combinatorial optimization problems. Based on the grouping optimization strategy, we designed a grouping optimization algorithm for natural image matting according to the assumption that the alpha matting within the group is similar. In the algorithm, the color average of the intra-group is selected as the representative of the optimization target, return the alpha matte corresponding to the optimization target in place of the group alpha matte. To verify the effectiveness of the proposed algorithm, the experiments study on alpha matting benchmark data set. Experimental results reveal that the proposed algorithm with a grouping strategy can significantly improve the alpha matte of the natural images.