Research Of Facial Landmark Location And Face Recognition Via AAM

Face recognition aims to obtain effective visual feature information from the face imagefor the identity authentication or recognition. As for the rising concerns on public informationsecurity, the requirements on identity authentication, and the need for face analysis andmodeling techniques in multimedia and entertainment，automatic face recognition has becomea hot research direction and has being widely used in public security and some core areas suchas human-computer interaction. The engineering research on face recognition began in the1960s, and has obtained a great progress after the development of nearly50years, but it stillfaces some challenges under unconstrained conditions such as uneven illumination, thevariability in posture and facial expression, the low resolution of the identifying image，change of the age, random block,etc.Automatic face recognition system mainly includes face detection, facial feature pointslocation, facial feature extraction and classification as well as facial feature matching andrecognition four components. Facial feature points location is a key step in face recognitionsystem, the feature point location accuracy significantly affect the performance of facerecognition system. Active Appearance Model was proposed by T.F.Cootes et al and has beingwidely used in the field of facial feature points location, but the traditional AAM model has apoor performance in fitting process.Aims at the above problems, we first have an in-depth study of the active appearancemodel and establish AAM model on human face in this paper. The reverse combinationalgorithm is applied to the fitting part of traditional AAM model to improve the fittingperformance. Secondly, we use Gabor transform to extract local facial features and analyzethe most representative linear subspace face recognition algorithm Eigenface and Fisherfaceby simulating the both algorithm on the ORL and Yale face database. We propose andimplement a new type of face recognition system after having a deep analysis of the abovetheories. The specific scheme is first to establish fitting improved AAM model on human facefor the precise positioning of facial features and choose Gabor transform and Fisherfacealgorithm for facial feature extraction and recognition. Through the integration of the various modules, we build and implement a face recognition system that are robust to the changes ofillumination, pose, expression and shade. Test the recognition performance of our facerecognition system on the ORL and Yale face database, experiments show that our system hasa better recognition effect than the classic Eigenface and Fisherface algorithm.