Font Size: a A A

Fundamental Research On Practical Face Recognition System

Posted on:2010-07-08Degree:MasterType:Thesis
Country:ChinaCandidate:J W DaiFull Text:PDF
GTID:2178360278463046Subject:Control theory and control engineering
Abstract/Summary:PDF Full Text Request
Recent years, building automatic face recognition system has become a hot topic ofcomputer vision and pattern recognition. And some commercial face recognition systemshave been developed and applied in public and individual security. In general, an au-tomatic face recognition system is accomplished in four steps, i.e. face detection, faciallandmarks localization, facial feature selction and extraction, face recognition. Face de-tection determines whether or not there are any faces in the image or video sequence and,if present, acquires the location and extent of each face. Facial landmarks localizationobtains the location of salient feature points of face, i.e. eyes, nose, mouth etc. Facialfeature selection and extraction acquires the most representative facial features. And facerecognition identifies or verifies one or more persons in the scene using a stored databaseof faces.In this thesis, some key issues are primarily studied, aiming at building real-timepractical face recognition system. And the main work of this thesis can be described asfollows:(1) Proposed a novel projection peak analysis method for rapid eye localizationThe projection peak analysis method achieves rapid and accurate eye localization bymaking use of the static rules of human face on the basis of uncomplicated computation.In order to eliminate the interferences (i.e. hair, eyebrow, glasses) around eye region, weimprove the general projection method by projection peak analysis. Experimental resultsshow that our method is e?ective, accurate and rapid in eye localization, especially whenthe face poses, illuminations, expressions, and accessories varied. Owing to the lowercomputation cost, our method can satisfy the requirement of real-time face recognitionsystem well.(2) Improved the face recognition method based on Gabor featureFor the high computational cost and long time consuming shortcoming of two-dimension convolution in the procedure of Gabor feature extraction, we proposed SFFS-Gabor face recognition method based on feature selection. In this method, through thelearning on training set by means of sequential ?oating forward search(SFFS), the mostrepresentative Gabor features with di?erent position, scale and orientation could be se-lected after two-stage procedure. Then the face description under the selected featurescould be employed in face recognition. Experiment result shows that this method couldensures high face recognition accuracy, and at the same time, it could reduce the com-plexity of algorithm and shorten recognition time evidently. After feature selection, the face recognition method on Gabor feature is more suitable for real-time practical facerecognition system.(3) Improved the face recognition method based on LBP featureThe dimension of general LBP feature is very high, so it is not suitable for real-timeapplication. To solve this problem, we reduce the feature dimension by taking advantageof compressed histogram. Experimental result shows that compressed-histogram methodcould achieve high recognition rate with short time consuming. In addition, we alsoproposed a novel LBP method based on total variation model(TVM-LBP). Total variationmodel could enhance the salient facial feature such as edges, profiles, corners and points,while LBP operator is sensitive to some local features, i.e. edges and corners of the image.The TVM-LBP method combines the advantages of TVM and LBP mentioned above,which could improve the face recognition performance, especially when the environmentalillumination has been changed.(4) Initially proposed and realized an open face recognition frameworkIn order to provide a reference prototype for the design of practical face recognitionsystem, we proposed an open face recognition framework(OFRF). Aiming at the universal,open and distributive properties, we studied the design pattern and holistic structure ofOFRF, and implemented the framework, depending on the support of three technicalplatforms, i.e. object-oriented technology, database technology and open computer visionlibrary(OpenCV).(5) Implemented a face recognition system for checking attendanceFor the purpose of daily checking attendance in o?ce environment, we implementeda real-time face recognition system using SFFS-Gabor method based on the open facerecognition framework(OFRF). This checking attendance system proved the open anduniversal properties of open face recognition framework and the e?ectiveness of SFFS-Gabor face recognition method in real-time practical application.
Keywords/Search Tags:Face Recognition, Rapid Eye Localization, Projection Peak Analysis, Gabor Wavelet, Feature Selection, Sequential Floating Forward Search(SFFS), Local Binary Pattern(LBP), Compressed Histogram, Total Variation Model(TVM)
PDF Full Text Request
Related items