Nonconvex γ-norm and Laplacian scale mixture with salient map for moving object detection

Moving object detection which has attracted wide attention is the critical issue of computer vision. Consequently, the low-rank and sparse decomposition (LRSD) has been a powerful technology for extracting the moving object from videos which has achieved improvement for moving object detection. However, it still has some defaults such as the lower degree for approximating the low-rank and sparsity components, ignoring the spatial information of videos, being sensitive to noise, and so on. To address these problems mentioned above, we propose a new LRSD method which is named nonconvex norm and Laplacian scale mixture with salient map (NNLSMSM). It adopts the nonconvex gamma-norm and the Laplacian scale mixture (LSM) to approximate the low-rank and sparsity components of traditional LRSD model for enhancing the degree of approximating. Meanwhile, a salient map mechanism which can effectively capture the spatial information from videos is introduced to NNLSMSM. In addition, we extend our proposed NNLSMSM method to a robust NNLSMSM (RNNLSMSM) method for enhancing its robustness via introducing a noise item. It can effectively solve the problem of being sensitive to noise. In addition, we adopt the alternating direction method of multipliers (ADMM) to solve our proposed NNLSMSM and RNNLSMSM methods. At last, extensive experiments which are performed on various popular datasets by some state-of-the-art methods demonstrate the effectiveness and superiority of our proposed NNLSMSM and RNNLSMSM methods.