Face Recognition Using Different Classifier Fusion Approaches

摘要

In this paper, three well-known feature extraction methods including BDPCA, PCA and LDA have been applied to visible and infrared images and the nearest neighbor method has been used to classify faces.After classification of extracted features, Dempster-Shafer, Fuzzy Integral and Decision Template methods are used to fuse the results of infrared and visible images distinctly. Considering the fact that infrared and visible images have their complementary features, a better recognition rate can be achieved by simultaneous use of these features. The complementary approach uses the three mentioned fusion techniques to combine the results of classifiers which have been trained using extracted features of infrared and visible corresponding images utilizing BDPCA method. The experimental results show that among three fusion schemes, best result is achieved by combination of classifiers related to complementary approach using Fuzzy Integral, while other fusion schemes have also improved the final recognition rate.