Co-saliency detection via cluster-based structured matrix decomposition
Zhengyi Liu, Song Xin Shi, Quntao Duan · 2019
Aiming at automatically discovering the common objects among a group of relevant and similar images as foreground, co-saliency has become a hot topic in recent years. Previous works utilize low-rank matrix recovery on the single image, but neglect the relationship between a set of images. In this paper, we propose a novel framework to capture the coherence of common salient objects, and solve the problem when the background is clatter. The model include a novel cluster-based tree-structured sparsity-including regularization that make regions from same class have identical saliency value, and a Laplacian constraint regularization is also integrated into the model, the propose is to enlarge the gaps between common objects and background in original feature space and smooth the saliency value in same cluster. Furthermore, to facilitate the efficient, a coherence weight is identified and integrated into the model. Experiment results on three benchmark datasets demonstrate are the performance of our method compared to other stateof-the-art co-saliency models.