Graph-based Poisson learning for image co-segmentation

Abstract

This paper focuses on applying the state-of-the-art graph-based semi-supervised learning algorithm Poisson Learning to the task of image co-segmentation. The weighted matrix built based on the image texture is used as the feature to do the classification by Poisson learning. The results presented in the paper show that Poisson learning does a good job in segmentation and has high accuracy at very low label rates. Moreover, Poisson learning is tolerant with multi-weight metrics and simple to implement.