Study On Face Recognition Based On Subspace Analysis Feature Extraction

In recent years, face recognition has become the most challenging researchsubject in the field of computer science and information technology, it has a veryimportant research significance and practical value. Like other biometrics of body, faceis born with people, good characteristics of its uniqueness and not easy to be copied foridentification provides the necessary premise. Compared with other types of biometric,face recognition has the non-mandatory, non-contact, concurrency, results the intuitiveand so on. Face recognition system mainly includes four parts, respectively is: thehuman face image acquisition, image preprocessing, face image detection, face featureextraction, matching and recognition. Feature extraction is a key factor in facerecognition，affect the performance of face recognition systems.This paper mainly aimed at the feature extraction algorithm and the correspondingexperiments, specific work can be summarized as follows:⑴Expound the face recognition technology and basic process of face recognition,Summarize background and realistic significance of face recognition. Outlined based onsubspace analysis, manifold learning algorithm of face recognition technology andResearch trends and progress in the field of face recognition.⑵The second chapter mainly elaborates the principal component analysis, theK-L transformation theory of one dimensional principal component analysis isintroduced in detail, analyzes the theory of principal component analysis essentially，getthe following conclusions: One dimension algorithm firstly converts face image into aone dimensional vector, damage the original internal structure of the face image,affecting the recognition rate. The corresponding two-dimensional algorithm is appliedto two-dimensional face matrix directly, can better keep the image of the originalstructure, but the features are statistical correlation. So the minimum correlationanalysis is introduced.⑶The main purpose of the linear discriminant analysis is to seek the bestidentification characteristic vectors. The third chapter introduces the theory of lineardiscriminant analysis and development. Compared with principal component analysis inthe second chapter, linear discriminant analysis is to extract the characteristics of betterclassification. The traditional one-dimensional linear discriminant analysis thatcommonly suffer from singular value problem in face recognition process.Therefore, the introduction of the two dimensional linear discriminant analysis algorithm to solvethe problem of singular values.⑷This paper finds Locality preserving projection method without consideringglobal information of data, and extracting discriminant features is statistical correlationbetween component by studying the Locality preserving projection. Therefore, putsforward the improved two-dimensional Locality preserving projection: on the basis ofthe rule of big span, introduce the minimum correlation analysis. Finally, in the threeface database to verify this algorithm respectively.