Research On Salient Object Detection Via Matrix Decomposition Algorithm Based On Non-convex Low-rank

Salient object detection is not ubiquitous but also challenging tasks in the study of computer vision.In this paper,we focus on the salient object detection models based on non-convex low-rank matrix decomposition.The main work can be summarized as follows:1.The basic problems about salient object detection are given,some basic mathematical definitions and theories of optimization used in modeling are discussed and some popular datasets about salient object detection and evaluation metrics are introduced.2.A novel regularization model for salient object detection is proposed,which integrates a weighted group sparsity with the convex Schatten-1 or the non-convex Schatten-2/3 and Schatten-1/2 norm,respectively.The corresponding alternative direction method of multiplier?ADMM?with derived solutions are discussed in detail,and the convergence of algorithm is validated.3.A new approach for the salient object detection is developed,in which the Schatten-2/3 is integrated with the non-convex sparse₂l₃norm.The proposed model essentially can be viewed as“Frobenius/nuclear hybrid norm+non-convexl₂₃ norm”,which can be set by splitting the objective function and then solved by using the alternating direction method of multiplier?ADMM?.Convergence of the algorithm is discussed in detail.And experimental results verify the efficiency of the algorithm.