| Face recognition has been becoming a biometric approach and popular issue in the field of pattern recognition with high scientific value and practical significance recent years because of its natural, direct and friendly trick. Face recognition involves many scientific fields such as machine vision, image processing, neural networks and mathematical calculation. Various pivotal algorithms and application technologies are growing to be mature. At present, the performance of face recognition has been improved continuously. The implementation of face recognition includes three stages which are face detection, facial feature extraction and the final classification. This paper is major in the second phase, researching and summarizing the algorithms for facial feature description, as well as some efficient facial feature dimension reduction.Feature description is the key point of the face recognition process. The quality of the Facial feature description mainly determines the result of the whole face recognition. How to extract the robust and discriminative features from the facial regions acquired at the first stage is the main task of the second phase. A new learning image filter can improve the performance of facial feature description through the linear image filtering before feature extraction. This article combines the linear filter with kernel trick, taking the first-order and second-order information of pixel gray level into consideration, proposing a new nonlinear learning filter which exceeds the linear one and the traditional handcraft-designed algorithms.The dimension of facial feature data is generally enormous which can cause the computational complexity and more error. In order to obtain the low-dimensional data for efficient representation of facial features, we should use effective dimension reduction methods to streamline complex vectors to brief ones which can better reflect the nature characteristics and classification structure of the original sample data. This paper presents the results of a comparative analysis of some popular dimensionality reduction algorithms.The main work and innovation of this paper:1. Research on the current popular facial feature descriptors and feature dimensionality reduction methods and taking all of them into experiments. Then, summarizing the theory of these algorithms and Comparing the performance of them with each other.2. Research on a kind of kernel-combined learnable filter assisting the face feature description. This discriminative filter considers the first-order and second-order differential information of pixel gray level to improve the performance of facial feature descriptors, addressing the common nonlinear problems encountered in face recognition. |
