A BAYESIAN FRAMEWORK FOR FACE RECOGNITION

作者：Daliri Mohammad Reza^*; Saraf Morteza

来源：International Journal of Innovative Computing Information and Control, 2012, 8(7A): 4591-4603.

• Bayesian decision rule
• K-nearest neighbor
• Parzen estimation
• Principle component analysis
• Support vector machine

In this paper, a statistical face recognition scheme proposed by combining the techniques of Bayes' theorem and Parzen estimation applied on various features such as discrete wavelet transform (DWT), Discrete Cosine Transform (DCT) and Principle Component Analysis (PCA). Parzen algorithm estimates the conditional probabilities for each class and according to Bayes' theorem; the class with maximum posterior probability is selected for each test face image. The optimal Gaussian variances for each class have been found by the Genetic Algorithm. (GA) optimization. The experiments on the ORL dataset demonstrate that the proposed Parzen based Bayesian classification method with enough DWT features leads, in mean recognition improvement, to 0.2% in comparison with Support Vector Machine (SVM) and 5.6% in comparison with K-Nearest Neighbour (KNN) classifier. Also applying various classifiers on DWT, DCT and PCA features, determine that with enough features of DWT, it has the best performance in compared with the others. Extra work has been performed to develop the statistical data dependence features selection in order to improve the recognition rate. This processed by searching in features space in order to minimize the reverse scattering matrix utilizing the GA. Results show that it significantly decreases the implementation complexity with selection of robust and informative features.

• 出版日期2012-7