.png){kind=icon}
.png){kind=icon}
Return
Local spatial continuity steered sparse representation for occluded face recognition
.png){kind=icon}
PRE
AI
DOI：10.1007/s11042-022-12427-9.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Recently, sparse representation in face recognition has been widely studied in computer vision. For face identification under complex conditions, many robust variants of sparse methods are proposed and achieved good results. However, the occlusion problem is still a challenging problem. Most of the state-of-the-art methods merely use pixel-wise error coding or structural error coding, which ignore the structural correlations between pixels. One can observe that occlusion has local spatial continuity. To make use of this local spatial continuity, this paper proposes a novel method, namely local spatial continuity steered sparse representation (LSCSR) for face recognition. The LSCSR uses a two-step strategy to calculate the occlusion support map. In the first step, we make use of information on residual to construct a matching mark image, and in the second step, we utilize the prior knowledge of the local spatial continuity and obtain the occlusion support map based on the matching mark image. Then, the weighted sparse coding framework is used to execute the face representation and classification. Extensive experiments on several public face databases demonstrate the effectiveness and robustness of the LSCSR in the face recognition against occlusion and other variations. Especially in AR dataset, we achieve the 4.00% performance gain in accuracy(96.50% vs. 92.50%) in Scarf occlusion.
Keywords:
Sparse representation
Face recognition
Robustness
Contiguous occlusion
Support map
Journal
IF:
3
Papers:
1.9W
Citations:
3.2W