The Research Of Human Face Recognition Algorithm Based On Symmetry

The uniqueness, acceptability, security, uniqueness and diversity of biologicalcharacteristics, compared with the traditional technology, making it have congenitaladvantage in authentication and recognition. Compared with other recognitiontechnology of human biological characteristics, face recognition has the advantage ofconcealment, non-contact, initiative, fast and intuitive, high accurate and reliable, andcost-effective, making it at national public security, civil, economic, familyentertainment and e-commerce and other fields has been widely applied. In recentyears, face recognition is a hot spot of concern in the field of information science andpattern recognition, the depth and breadth of its relevant technology continue todevelop. The face recognition technology and the research content involved withdigital image processing, computer vision, data mining, pattern recognition, machinelearning, artificial neural network, database, legal ethics and psychology, etc.Mirror symmetry is the more significant attributes of facial contour and facialorgans in face image, through the image transformation of face image, people canaccurately identify the face image after image transform. Therefore, the mirror imagecan also be used as the training sample of face recognition, so as to enlarge the samplesize of face recognition. In fact, it expands face recognition sample size by imagetransformation and provides a new and effective solution for the small sampleproblem of high dimensional space of face recognition. Therefore, the main work ofthis article is the research of the face recognition algorithm based on symmetry. Theresearch of the face recognition algorithm based on symmetry is the goal of thisarticle. On the basis of kernel principal component analysis and direct lineardiscrimination analysis, introduction of mirror symmetry and parity decompositionprinciple, combination of kernel principal component analysis and direct lineardiscrimination analysis, two new face recognition algorithm is proposed, and usingtwo different face database, we respectively validate the recognition performance ofthe two algorithms under different environment and conditions. The main researchwork and contributions of this paper are as follows:1. Proposed the symmetrical kernel principal component analysis of face recognition algorithm. Aiming at the problem that Kernel Principal ComponentAnalysis(KPCA) is not considering the symmetry of face evident features, andgenerally lacks of training samples in face recognition, recognition rate remains to befurther improved. Therefore, this paper introduction of mirror symmetrycharacteristics on the basis of Kernel Principal Component Analysis algorithm. Oddsymmetry samples and even symmetry samples are obtained by the image transformof face training sample. Feature component is extracted by odd symmetry samples andeven symmetrical samples. Finally, the nearest neighbor classifier is employed toclassify the extracted features. The algorithm is evaluated on the ORL face imagedatabase, Experimental results show that this algorithm multiple enlarges the numberof face training samples, achieved better performance than the kernel principalcomponent analysis algorithm.2. Proposed the symmetrical direct linear discrimination analysis of facerecognition algorithm. This algorithm on the basis of direct linear discriminationanalysis, combining more remarkable symmetry of face, the odd and even symmetricsample are obtained by use the odd-even decomposition principle and imagetransform for the face training sample, and the odd and even symmetrical principalcomponents are respectively extracted, the algorithm uses the minimum euclideandistance to classify the feature. This paper conducts experiments for this algorithm onthe ORL and YALE face image database, experimental results show this algorithm notonly multiple enlarges the number of training samples, but also the recognitionperformance is superior to the direct linear discriminant analysis algorithm.