Multiple feature fusion for classification of facial images

In this work, we demonstrate the utility of a set of features which are popular for object recognition, for classification of human faces. Face Images are represented comprehensively by the integration of complementary features such as Geometric Blur(GB), Local Self Similarity(LSS) and Pyramid of Histogram of Visual Words(PHoW). We also propose a novel feature fusion method to combine various visual features in a multiple kernel learning framework for automatic classification of face images. MKL algorithm is used to estimate optimal weights to combine image features and achieve superior performance in classification. Experimental results on face datasets like Yale, AR and Movie show that the fusion of multiple features can achieve higher classification accuracy than classifiers with single feature descriptor as well as compare favorably with several state-of-the-art approaches.