Face recognition based on singular value and feature-matrix

The face recognition algorithms based on singular value decomposition (SVD) have low recognition accuracy due to the common essential defect which singular value vector of arbitrary two face images have the different basis spaces in general. According to this, a weighted adaptive algorithm based on some important partial features is proposed. It normalizes different faces and then locates the features of eyes, nose and mouth with horizontal and vertical projections. Subsequently, local features of the key parts of face are extracted and weighted respectively by singular value to get the feature-matrix. Dynamic method of how to choose the weights of local features and formula of how to obtain the feature-matrix is given. Finally, the developed support vector machine is utilized to recognize faces. Experiments show that the proposed algorithm can not only calculate efficiently and work easily, but also deal with low recognition rate issues in SVD, which shows a good potential of application.