The Improved Kernel Principal Component Analysis Based On Virtual Samples And L1 Norm For Face Recognition In Small Samples

Unique advantages of face make face recognition become one of the most widely used authentication technologies in the field of biometric technology. It has gained rapid development and great practical significance. But in real life, it is difficult to get a man’s multiple face images, which are easily affected by posture, facial expressions and other nonlinear factors. if the sample size is less or even single and non-linear factors exit in face images, the extracted principal components by the traditional principal component analysis based on L2 norm and other dimensionality reduction algorithm does not guarantee better classification results. In this paper, how to create effectively the virtual sample to increase the sample size, how to weaken the influence of nonlinear factors such as gesture on face recognition and other issues are in-depth researched. The main work is as follows:1. In order to solve the problem that the number of training and testing samples is small in face recognition, we propose a method that can create virtual samples in category on the basis of the relevant research achievements at home and abroad. The main principle is divide face image in database into the front face image and non-front face image, and then different transformation will be conducted on those, which is not only help to add more different poses samples thus weakening the influence of nonlinear factors such as posture and others on face recognition, but also help to enhance the reliance of new virtual face sample on samples in the original face database. Ensuring the recognition rate, this method improves the stability and reliability of the recognition result.2. Questions and algorithms exist in the field of face recognition are in-depth researched. Then, kernel principal component analysis based on combinational creating virtual samples in category and L1 norm for face recognition is proposed and the corresponding basic algorithm is given. In a sense, this method can "multiply" weaken the influence of nonlinear factors such as gesture on face recognition for having better classification performance. We also make a combination of the method to combinational creating virtual samples in category with kernel principal component analysis, the method to combinational creating virtual samples in category and L1 norm principal component analysis, L1 norm principal component analysis and kernel principal component analysis methods, which are compared with the proposed method to prove its superiority.A lot of experiments are conducted respectively in face databases and the concept of recognition rate variance is proposed for discussion of that overall performance. Experimental results show that the recognition rate of the improved kernel principal component analysis based on creating virtual samples in category and L1 norm for face recognition in small sample are not only improved, but also the results are relatively stable and reliable.