Learning a Locality Preserving Subspace for Visual Recognition

• Authors: Xiaofei He;Shuicheng Yan;Yuxiao Hu;Hong-Jiang Zhang
• Venue: ICCV '03 Proceedings of the Ninth IEEE International Conference on Computer Vision - Volume 2
• Year: 2003

Previous works have demonstrated that the face recognitionperformance can be improved significantly in low dimensional linearsubspaces. Conventionally, principal component analysis (PCA) andlinear discriminant analysis (LDA) are considered effective inderiving such a face subspace. However, both of them effectivelysee only the Euclidean structure of face space. In this paper, wepropose a new approach to mapping face images into a sub-spaceobtained by Locality Preserving Projections (LPP) for faceanalysis. We call this Laplacian face approach. Different from PCAand LDA, LPP finds an embedding that preserves local information,and obtains a face space that best detects the essential manifoldstructure. In this way, the unwanted variations resulting fromchanges in lighting, facial expression, and pose may be eliminatedor reduced. We compare the proposed Laplacian face approach witheigenface and fisherface methods on three test datasets.Experimental results show that the proposed Laplacianface approachprovides a better representation and achieves lower error rates inface recognition.