Face Recognition Based on Null Space of MFA and Bidirectional Two-dimensional Algorithm Based on Divisor Method
|Course||Control Theory and Control Engineering|
|Keywords||Face Recognition Feature Extraction Null Space of Marginal Fisher Analysis An optimalFLD algorithm for facial feature extraction Bidirectional two-dimensional algorithm Divisor method|
Face recognition technology has become an important way of pattern recognition field. Due to itsspecial merits of flexibility, economy and accuracy, face recognition technology has a broad applicationfuture in biometrics security field. Face recognition can be divided into three steps as follows: facedetection, feature extraction and classification. As the high dimension, the image processing is verycomplex. So feature extraction becomes one of the key steps in face recognition. In this paper, based on theclassic feature extraction algorithm, we bring up face recognition based on null space of Marginal FisherAnalysis and Bidirectional two-dimensional algorithm based on Divisor method.The main research content and major work of this pater is as following:1. Based on Marginal Fisher Analysis (MFA) and optimal Fisher discriminant analysis algorithm, weput forward the null-space Marginal Fisher Analysis, which was extracted feature respectively using nullspace and the non-zero space, and solve the singularity problem of the within-class scatter matrix. In such away, the small sample size problem occurred in MFA is avoided. The experimental results on ORL andYALE face database show that the proposed method yields greater recognition accuracy and prove theeffectiveness of the algorithm;2. Under the framework of two-way two-dimensional algorithm, we use the divisor method. Thismethod calculates the percentage loss of mapping matrix in both row and column directions firstly, andthen use the Divisor method to select the numbers of two mapping matrices’ projection vectors, which baseon the principle of minimum total percentage loss. The experimental results on ORL and YALE face database show that the proposed method yields greater recognition accuracy while reduces the overallcomputational complexity.At the classification stage of the facial feature, the nearest neighbor classifier is used zero-spaceboundary Fisher analysis algorithm; Otherwise, two-dimensional nearest neighbor classifier. Simulationexperiments using ORL and Yale face database were made in this paper, and experimental results show thatthe proposed method has better recognition effect.