Orthogonal Complex Locality Preserving Projections Based on Image Space Metric for Finger-Knuckle-Print Recognition

Manifold structure is important for a data set, and many subspace learning methods tend to preserve this structure in the learning process. In this paper, we simultaneously consider distances and angles between image data vectors to measure data similarities, in hope of more sufficiently capturing the manifold structure. In order to highlight the distinctions among angles between different data, and enhance the complementary information of angles compared with distance, we propose a new type of image angle measurement in a shifted image space that centered at the data mean. Both angle and distance are fused using the parallel fusion strategy, based on which we propose the complex locality preserving projections (CLPP) to extract low dimensional features that can better preserve the manifold structure of input data set. In order to remove redundant information among features, we further extend CLPP to the orthogonal complex locality preserving projections (OCLPP) approach, which produces orthogonal basis functions. Experimental results on PolyU finger-knuckle-print database show the effectiveness of our proposed approaches, which achieve better recognition performance compared with related mainfold-preserving learning methods.