Face recognition with singular value decomposition based common matrix approach in one sample problem

Face recognition system used in the real applications such as information security, smart cards, credit card authentication, identification of criminals and etc., is one of the most used biometric recognition systems. An important subject in these kinds of applications taking attention of researches is one sample problem. One sample problem is the situation of being one sample per person in the training set. The methods using one sample in the training set for recognition have advantageous in the situations that collecting many samples per person is difficult and for storage requirements. However, many common face recognition methods will suffer serious performance drop or even fail to work when the computation of within-class scatter matrix is required.In this thesis, one sample problem is investigated and it is tried to develop an algorithm intended to use Common Matrix Approach, which has high recognition results and requires the computation of within-class scatter matrix, in the one sample situation. For this reason, an algorithm using singular value decomposition is developed in which the one sample for each class is projected on the vectors spanning the null space of the image and a common matrix is obtained for each class.In addition, a method called Combined Common Feature Subspace is presented to increase the recognition performance of this algorithm. In this method, global features are obtained by using whole face image and local features are obtained using eye, nose and mouth regions of the face in which the most pixel changes occur.Experimental results in the SVD based Common Matrix Approach show high recognition rates with same facial expressions and different illumination conditions. In addition, it is observed that Combined Common Feature Subspace Method presented to increase the performance of this algorithm increases the recognition rates especially when the images with different facial expressions are used.Key Words Face recognition, one sample problem, common matrix, singular value decomposition, subspace methods