Face Recognition Based On Manifold Learning And Support Vector Machine

Face with its natural, intuitive, safe, fast characteristics has become one of the mostpromising biometrics recognition technology. Its application in security systems, credit cardverification, video conferencing, public security have become a research hotspot in the fieldsof pattern recognition and artificial intelligence. However, the complexity of human facialstructure and the diversity of facial expression, exploration of effective feature extractionalgorithms and classifiers with strong generalization ability become the new challenge of facerecognition systems.Manifold learning as a new nonlinear dimensionality reduction method can effectivelyfind low-dimensional nonlinear manifold data embedding in high-dimensional space andmining hidden intrinsic information. Support Vector Machine (SVM) based on structural riskminimization principle, can find the best compromise between complexity and learning abilityof the model according to some effective sample information. And it has strong ability ofgeneralization.This paper combines manifold learning and SVM, applying to two modules for facerecognition-feature extraction and classification. Proposing two improved algorithm ofmanifold learning feature extraction. And using the firefly algorithm optimized parameters tosupport vector machine for simulation and the validity of the algorithm. The main contents ofthis paper as follows:(1) Proposing a new algorithm, which is based on local linear embedding of thedynamically determining neighborhood parameter. This algorithm can automatically determinethe neighborhood of the data point by using the single-chain clustering and furtheroptimization algorithm. Reducing the insufficient of Local Linear Embedding algorithm fixed neighborhood, as well as deal with the reality of non-uniform failures of source datasets.Combined with the optimized support vector machine for classification,the theoretical andexperimental results shows the high recognition rate of this algorithm.(2) Proposing a face recognition algorithm of semi-supervised locally linear embeddingwith improved distance. This algorithm introduces new distance rules, improving the highdemand for the construction of the sample when the sampling density neighborhood structurebased on Euclidean distance. Using the improved algorithm and Support vector machine(SVM) classifier for face recognition get a higher upgrade of the average recognition rate onthe ORL and YALB face database. And by analyzing the experimental results of ORL andYALB face database can indicate the effectiveness of this algorithm.