Face feature extraction based on reduced-dimension matrix of DCT and projection of block covariance matrix

Feature extraction is the foundation of face recognition and plays a very important role in face recognition. Different method of face feature extraction has different performance and efficiency to later steps of face recognition. This paper firstly discusses how to reduce dimensions of face image matrix obtained by DCT from video. Then we can acquire covariance matrix of block face matrix which consist of low frequency factor and high frequency factor. The low frequency factor retains basic and stability features of the same individual and high frequency reflects the details of face contour, etc. Lastly the eigenvalue is acquired through calculating the covariance matrix. The obtained eigenvalue is sorted and the projection of three biggest eigenvalue is taken as the key factors which reflect face block matrix to recognize face. The experimental results demonstrate that comprehensive utilization of high frequency and low frequency factor and big eigenvalue can keep more energy of image, so it can achieve high recognition efficiency.