An alpha matting algorithm based on collaborative swarm optimization for high-resolution images

Abstract

High-resolution image matting is one of the challenges in image composition and foreground extraction. It is essentially a large-scale combinatorial optimization problem for foreground/background pixel pairs. However, little attention has been paid to this issue. A multiclass collaborative optimization strategy based on RGB color clustering is proposed to reduce the dimension of this problem, addressing the issues caused by its ultrahigh dimension. This paper presents a collaborative feedback grouping strategy to solve this large-scale combinatorial optimization problem. Based on these two strategies, a competitive swarm optimization algorithm based on group collaboration (GC-CSO) is proposed. Its performance is verified experimentally by using an alpha matting dataset, showing that it can significantly reduce the dimension of the image matting problem and outperform the existent large-scale optimization algorithms with grouping strategies in the alpha matte comparison.