Study On Improved Algorithms Of Face Verification

The technology of face verification is a hot topic in the area of pattern recognition. It contains three main steps: face image preprocessing, feature extraction and classifier design. Face feature extraction is also referred to as face representation. The goal is to model or represent high dimension face patterns in low dimension feature space so as to extract face feature with discriminant power for classification. Feature extraction is a key step for face verification and also as the primary difficulty in face verification, so it is a hotspot in recent studies.The thesis studies the theories and methods of feature extraction in face verification, focusing on the Client Specific discriminant analysis. The main points are as follows:(1) An improved face verification algorithm is proposed based on the combination of modular 2DPCA and CSLDA in this paper. Feature extraction of Client Specific Linear Discriminant Analysis (CSLDA) transforms an image matrix to a vector which caused great dimensionality and computational complexity. Furthermore, the local feature is not considered in CSLDA. Then the new method is studied to avoid the deficiency. The initial features are extracted with the original images which are divided into modular sub-images. The 2DPCA is performed to get the low dimensional features which can be computed conveniently. The local features are extracted efficiently using the proposed new method. Then CSLDA is utilized on the new pattern which is obtained through the modular 2DPCA to extract the final features. Contrasting with PCA, the discriminant information obtained from the between-class scatter matrix and within-class scatter matrix are included using CSLDA. Moreover, client specific subspace could describe the diversity of the different face better and has more robust discriminant information than the traditional LDA. The experimental results obtained on the facial database show that the verification performance of the new method is superior to that of the primary method CSLDA.(2) By the induction of the (1), an improved face verification algorithm is proposed based on the modular 2DPCA and CSKDA which introduce kernel operation into the linear discriminant analysis method. The experimental results obtained on the database show the effectiveness of the proposed method.(3) This paper researched the kernel adaptive learning algorithm. The kernel function is introduced to solve the nonlinear pattern recognition problem. The advantage of a kernel method often depends critically on a proper choice of the kernel function. A promising approach is to learn the kernel from data automatically. Over the past few years, some methods which have been proposed to learn the kernel have some limitations: learning the parameters of some prespecified kernel function and so on. In this paper, the nonlinear face verification using learning the kernel matrix is proposed. A new criterion is used in the new algorithm to avoid inverting the possibly singular within-class which is a computational problem. The experimental results obtained on the facial database show that the verification performance of the new method is superior to that of the primary method Client Specific Kernel Discriminant Analysis (CSKDA). The method CSKDA need to choose a proper kernel function through many experiments, while the new method could learn the kernel from data automatically which could save a lot of time and have the robust performance.