Face Recognition Algorithm Based On Multi-pattern Fusion

Face recognition as a friendly, non-contact biometric recognition technology, can be divided into two-dimensional face recognition and three-dimensional face recognition. Studies of the two-dimensional face recognition based on image were carried out earlier. In the laboratory environment, methods of two-dimensional face recognition can receive a high recognition rate. However, FRGC (Face Recognition Grand Challenge) tests show that even the best two-dimensional face recognition algorithm cannot provide reliable identification results when illumination, posture, facial expression change violently. The three-dimensional face recognition can be overcome or mitigate the impact of these factors. By applying the information fusion theory, integrate the two-dimensional face recognition algorithm and three-dimensional face recognition algorithm in decision level can get efficient and stable recognition results.In order to solve the problem of that the two-dimensional Gabor feature dimension is too large, this paper proposes the R-2DPCA algorithm by improving 2DPCA (two-dimensional principal component analysis) method. Formed a complete two-dimensional face recognition algorithm based on Gabor+R-2DPCA. Adopt the neighbor classifier respectively in CASIA face database and on ORL face database, experimental results show that the Gabor+ R-2DPCA algorithm has better performance than PCA and 2DPCA and the highest recognition rate reached 97.5%.This paper discussed conversion process of transforming the three-dimensional model to the depth images and proposed LTP (local ternary patterns) operator which is suitable for feature extraction from depth images. Without significant increase in algorithm complexity, the LTP method showed good noise immunity. Achieved Fisherface method and constituted a three-dimensional face recognition module by combining Fisherface method and LTP method. The result from the experiment based on CASIA three-dimensional face database shows that the proposed three-dimensional face recognition algorithm is feasible. The highest recognition rate reached 95.8%, the training time and the test time compared to the Fisherface method increased less than 4%.Based on the information fusion theory, the paper proposed four fusion models between 2D images and 3D models. By analyzed the four models, we proposed a face recognition model fusing in decision level. In this model, obtained N candidates through Gabor+R-2DPCA face recognition algorithm, then use depth image LTP+Fisherface algorithm to get the final recognition result. Then we designed the multi-modal face recognition system software according to the proposed model. Through the experiment, when the N value greater than or equal 30, the system recognition rate stable at 98.5% above.With the development of three-dimensional imaging technology, the face recognition algorithm of fusing two-dimensional images and three-dimensional model proposed in this paper will have broad application prospects.