F-2DCCA: A New Fuzzy Feature Extraction Method for Face Recognition

摘要

Integrating the sample distribution information into the process of feature extraction is beneficial to classification accuracy. In this paper, a fuzzy two-dimensional canonical correlation analysis (F-2DCCA) method is proposed for image feature extraction . By making use of the Fourier transform and fuzzy k -nearest neighbor algorithm, we first construct a new fuzzy class-membership matrix to represent the distribution of image samples. Furthermore, two improvements based on Two-dimensional Canonical Correlation Analysis (2DCCA) are presented to promote the discrimination performance of the feature vectors and reduce their dimension respectively. The experimental results on the combined face database demonstrate the feasibility and effectiveness of the proposed approach.