Abstract
Salient object detection is not only important but also challenging tasks in the study of computer vision. In this paper, different from existing approaches, we propose a novel regularization model for the salient object detection, which integrates a weighted group sparsity with the convex Schatten-1 or the non-convex Schatten-2/3 and Schatten-1/2 norm, respectively. A weighted group sparsity induced norm developed in this paper is shown to be able to make the foreground being consistent within the same image patches by effectively absorbing the image geometrical structure as well as the feature similarity. The Schatten quasi-norm is successfully used to capture the lower rank of background via factorization technique, and an alternative non-convex formulation for nuclear norm is studied. The corresponding alternative direction method of multiplier (ADMM) with derived solutions are discussed in detail, and the convergence of algorithm is validated. Extensive experiments on the six widely used datasets show that the proposed approach has capacity in outperforming most of state-of-the-art models in current literature.
Accepted Version (
Free)
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call