Research On Face Recognition Using Hidden Markov

Face recognition is one of the most challenging problems in the fields of patternrecognition and machine vision. It has a wide range of promising application, such ashuman-computer interaction, security surveillance, identity authentication, and so on.As a statistical method, Hidden Markov Model (HMM) can describe and recognize thehuman face by combining the numerical vectors of its various organs. Therefore, goodresults can be achieved.In this dissertation, some improvements are made on the shortcomings of featureextraction of HMM firstly. More robust feature vectors can be extracted from theoriginal images, using the Principal Component Analysis (PCA), Nonnegative MatrixFactorization (NMF), 2D Discrete Cosine Transform (DCT) and Singular ValueDecomposition (SVD) respectively. And these vectors can be utilized as the observationvectors for HMM training and recognition. Experimental results show that the proposedmethods can not only improve the recognition rate, but also increase the efficiency ofrecognition significantly.Secondly, HMM method is improved from the perspective of information fusion.With the traditional HMM approach, a highly precise face model can be built using theobservation vectors generated by overlapping technique. However, it costs more timefor training and identifying. Feature level fusion and decision level redressal are used toimprove the traditional HMM on the condition of no-overlapping technique. Twodifferent vectors are combined by the Canonical Correlation Analysis (CCA) in theFeature level, and the fused-vectors are used as the observation vectors for HMM. Andin another approach, the distance decision is introduced into the HMM frame, whichcombines the minimum distance with the posterior probability to, enhance theperformance of the classifier based on HMM. Compared with the traditional HMMmethod, the novel approaches can keep the highly recognition rate, and reduce muchtime of HMM training and recognition synchronously.