The Research On Face Recognition Algorithms Based On Elasitc Garphic Matching And Hidden Markov Model

With the development of modern science and technology, the biometric featureidentification technology has been researched by many academics and institutes inrecent years. Meanwhile the products researched and developed are widely used inmany fields such as information security, financial business, social security, importand export management, administration, traffic and medical care system.The most applicable technologies of biometric feature identification are basedon face, iris, finger and palm mark, ear and hand figure. Because of the advantagesof its untouchable, easy and convenient character, the Face Recognition technologyhas become the mainstream object to research in biometric feature identificationfield. However, face images are usually affected by light, angle, age andenvironment, and different face structures have high comparability, so how toprocess face recognition quickly and accurately, is a necessary problem to solvewhen researching the face recognition technology. The major contributions of thisthesis are summarized as follow:1.This thesis summarizes all popular methods in face recognition field, andintroduces their theory analysis, approach to realization and representationalresearch contribution, then analyses their advantages and disadvantages in facerecognition process briefly. Single method or feature is always limited by someenvironmental conditions, so the fuse of different methods and features can avoid theflaws and enhance the robust character.2.This thesis presents a new method with optimized wavelet weight to theGabor wavelet parameters, as a solution to the diversity problem caused by differentfrequency eigenvector in Elastic Bunch Graphic Matching (EBGM) Algorithm offrontal face recognition. According to the various influencing degrees of the waveletparameters, classify them into several parts and give each part different weight, sothat the diversity of different face images is enlarged evidently.3.According to the research and analysis to the face recognition system basedon Hidden Markov Model, this thesis presents a method to build a model withsingular value decomposition. Because of the stability, displacement immutability,transposition immutability and proportional change with the image luminancecharacter, this model has certain anti-jamming to the change of light, angle and environment. In comparison with the model using gray value or2D discrete cosinetransform directly, the Hidden Markov Model based on singular value can achieve tohigher recognition rate.