Study On The Single Face Image Recognition Based On Integrated Learning Model

Single sample face image recognition means that in the case which every class training has only one picture, to determine the type of test samples. As an important research direction of face recognition, a single sample of human face recognition is widely used in transit inspection, identification of suspects and other occasions. Because the sample is too small, and by the impact of unfavorable factors light, rotation, how to improve the accuracy of single samples of human face recognition is a more difficult problem. To solve this problem, we have carried out work in the following two aspects:1, We present a based on local generic representation with PSO (Particle Swarm Optimization:PSO) single sample face image recognition method. In our method, first, using a common set of training images in lighting, occlusion by the human face of changes in experimental comparison and generates changes in the dictionary. Second, the various parameters of query sample partial represented by PSO optimization, including the block size, the neighborhood size, the size of the overlap and regularization parameters. Finally, the query samples by parameters optimized for local representation and query samples represented by the dictionary of variety and training samples is identified. Since PSO optimization algorithm can find the optimal solution for a given parameter, so this article will use the parameters optimized to improve the recognition rate.2, We present a single sample of human face recognition method based on Gist features, In our method, first, on the face image pre-processing, while using Gabor filter image is filtered to remove the effects of light on the image. Secondly, we extract Gist feature on face images. Finally, KNN classifier to classify the extracted features, and get the recognition rate. Since Gabor feature not sensitive to light, and Gist is characterized in a certain area of the average characteristics of the taken direction and scale, so the proposed method of face images in light and rotation, etc. has a good effect.Two methods are proposed in the plurality of sets of face database experiment. Experimental results show that the proposed algorithm effectively improves the current method has some key issues, and to further improve the recognition rate of single training sample.