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.
Mendeley helps you to discover research relevant for your work.
CITATION STYLE
Feng, F., Huang, H., Wu, Q., Ling, X., Liang, Y., & Cai, Z. (2020). An alpha matting algorithm based on collaborative swarm optimization for high-resolution images. Scientia Sinica Informationis, 50(3), 424–437. https://doi.org/10.1360/SSI-2019-0181