The Research On Facial Feature Points Alignment Technology Based On Shape Model

Automatic Face Recognition (AFR) technology is trying to endow the computer with the capacity of identifying human identity according to facial features.The study has important scientific significance and enormous applied value. After thirty years of development, AFR technology has made considerable progress.At present, we have obtained acceptable recognition performance using the best face recognition system under ideal conditions. However, testing and practical experience shows that face recognition technology under the non-ideal conditions is still far from ripe and it is necessary to resolve many key issues to develop a really robust, practical application of AFR system. In particular,we need to study precise alignment of key facial features as a necessary precondition for recognition. Feature points alignment far from being a problem has been solved. In particular,it required that we should attent when we solved the problem under non-ideal conditions. Therefore, in order to develop a truly practical face recognition system, we must give full attention to researchs on the key facial feature points alignment. In this paper,the main research work is as follows.This article focuses on the subject of facial feature points alignment based on statistical study. We present the basic principles of ASM/BTSM detailedly , introduce the ASM/BTSM algorithm based on the appearance of gray-scale model and introduce the expansion of ASM/BTSM model systemly.We analysis the deformation model, iterative search and the matching based on the appearance of gray-scale of the ASM/BTSM.An algorithm of shape model alignment based on local binary pattern is proposed in this paper. The algorithm extracts LBP block which is the center of the rectangular area of feature points as the local feature of facial feature points, establishs a distribution model based on reconstruction error using principal component analysis algorithm,and then searchs the shape to be matched iteratively. In addition, in order to reduce the error caused by points to be matched are not on the direction of normal,we use the mathod searching the best outline points on eight direction.We compare the advantage and disadvantage with the algorithm in this paper and the traditional algorithm in experiment. Experiment results show that LBP-BTSM algorithm is more precise than the traditional algorithm.But pose is still the main question of feature points alignment. An algorithm of weighted Bayesian tangent shape model(WBTSM) based on shape evaluation is presented in this paper.We define a function based on shape evaluation in the basic of local texture model, which measure the extent to be matched with the shape to be searthed and training data. WBTSM algorithm using the information of shape evaluation,projects the shape being searched to the shape subspace in the manner of weighted,which is unlike the Bayesian tangent shape model using orthogonal projection. Compared with the orthogonal projection, the weighted projection can use the information in the progress of searching,and the search may jump out of local extremum,so we can get more accurate results.Finally,we design a windows-based automatic face recognition system that can extract facial feature effectively. The system includes three main components of face recognition which are face detection,feature location and face recognition.