Recognition Of Nuclear-based Methods

Human face is our primary focus of attention in social intercourse, playing a major role in conveying identity and emotion. Although human can recognize face and its expression easily, automatic face recognition is till a very difficult problem for computer. On the other hand, due to the face recognition technology has a broad range of potential applications in video surveillance, access control, personal card identification, multimedia database retrieval and security, etc, face recognition is one of most active research subject in the area of pattern recognition and computer vision.Facial feature extraction and classifier are two key problems in face recognition. We applied the kernel methods which proposed in pattern recognition recently to deal with above two problems and got some useful results.The main contributions of this thesis are as follows:1, A kernel function based on fractional inner-product is presented. First, we discuss two kinds of kernel function based on distance and inner-product, and the properties that a kernel function for classification should possess are: (1)The correlation which is strong between samples in primary space should turn to weak after projecting to feature space;(2). The correlation which is weak between samples in primary space could maintain the weak correlation in feature space. Base on these, we propose a kernel function include fractional inner-product model which is better fulfill these properties, and apply it to kernel principle component analysis. The experiment results based on AR and ORL face database show that the KPCA method include fractional inner-product function is superior to KPCA method based on polynomial kernel function in terms of recognition accuracy and stability to the variations between the images of the same face due to illumination expression and viewing direction.2, This paper presents a novel KPCA+Null Space method by integrating the kernel PCA and the null space of the within-class scatter matrix. The kernel PCA method which extends to include fractional inner-product models first derives nonlinear features of face samples, then we construct the null space of the within-class scatter matrix, and calculate the optimal discriminant vectors by maximize the between-class distribution, after the projection of the samples onto the optimal discriminant vectors, we can obtain the optimal discriminant feature vectors. Our test results based on AR and...