Using Particle Swarm Large-scale Optimization to Improve Sampling-based Image Matting
Liang Lv, Han Pang Huang, Zhaoquan Cai, Hu Hui · 2015
Sampling-based image matting is an important basic operator of image processing. The matting results are depended on the quality of sample selection. The sample selection produces a pair of samples for each pixel to detect whether the pixel is in the foreground of an image. Therefore, how to optimize the production is usually modeled as a large-scale optimization problem. In this study, particle swarm optimization is applied to solve the problem because its property of rapid convergence is positive to the real-time demand of image matting. We regard every two dimensions of a particle as a sample pair for a undetermined pixel. The encoding can make image matting more effective when there are relevant pixels in the image. The experimental result indicates that the proposed particle swarm optimization performs better than existing optimization method for image matting.