Face Recognition Based On Local Feature And Evolutionary Algorithm

Face recognition is a kind of biometric recognition method. Comparedwith other biometric recognition methods, it has unique advantages, andhas arisen worldwide concerns. Face recognition research can enhance theawareness of human thinking model and processing of visual information.Finally the research can form the computational models to simulate humanmind, which can recognize the faces and identify objects.The face biological characteristics determine the complexity anddiversity of facial features. The external factors and the user behaviorscan affect the recognition accuracy. The interference factors includeillumination, posture, expression, blocking, age. The impact of thesefactors can make the same person face images different, when these faceimages are collected in different time or in different place. The localfeatures influence less, and can describe the details of face. But theposition of local features are changed, the matching processing of localfeatures becomes very complex and difficult, but the correct matching oflocal features is the key to recognize the faces based on method of localfeature.The processing of facial feature points matching in two images isactually calculate the topology of an image features to fit to the otherimage in best-coverage mode. It makes the matching processing becomes anoptimization problem, there are many effective solutions of optimizationproblem, this paper focuses on evolutionary algorithm.The evolutionary algorithm relies on alternating the global searchprocess and local search process to solve optimization problem. The globalsearch process accomplish the global constraint of local features, andthe local search process accomplish the local feature optimization andjump out from local extremum. This method is very suitable for the methodof face recognition based on local feature, the two processes alternately execute to finish the facial feature matching in best-coverage mode, andthen calculate the similarities of matching features to recognize thefacial images.For the interference of posture factors, the rotated featureclassifier was proposed to detect the rotation face and calculate therotated angle, the angle was used to normalize the rotated face image.First, the rotated feature was constituted based on LBP, and the rotationangle value was added. Second, with this feature, the rotated classifiersand regular classifiers were trained by AdaBoost method. The rotatedclassifiers were used to detect the rotated face in image, and the regularclassifiers were used to verify the result. A new principal directionmethod was proposed to calculate rotation angle of facial image with highprecision. The experiment results indicate that the new method can detectthe face under all degree rotation in image plane with high speed andnormalize the rotated facial image, the performance is better than otheralgorithms.For the interference of illumination factors, the local feature ofillumination invariance was used to extract facial features. The SIFT(Scale Invariant Feature Transform) feature has scale invariance, affineinvariance and rotational invariance, and the feature is pixel gradient,so it has illumination invariance. The SURF (Speeded Up Robust Features)feature is approximated version of SIFT feature, it also has theadvantages of SIFT feature. The experiment results indicate the SIFT hasa good performance in face recognition, the SURF feature is faster thanSIFT features.For the interference of expression factors, the PSO (Particle SwarmOptimization) method of features matching based on evolution algorithmwas proposed. The evolution process of PSO algorithm was improved, theface biological structure is the global constraint, the features matchingwas formed in best-coverage mode. The matching features were used tocalculate the similarity, and recognize the facial image. The experimentresults indicate the method based local feature and evolution algorithmcan recognize the face with high precision. For the interference of blocking factors, the MEA (Mind EvolutionaryAlgorithm) method of the blocking and non-blocking feature classificationwas proposed. The non-blocking features were used to recognize theblocking facial image. The blocking features and non-blocking featureshave regional and continuity, these properties can be used as the globalconstraint, the evolution process of MEA algorithm was improved toclassify the two kinds of features. The experiment results indicate themethod based local feature and MEA can recognize the blocking facialimage.For the interference of age factors, the SFLA (Shuffled Frog-LeapingAlgorithm) method of the long time span feature and short time span featureclassification was proposed. The long time span invariance features wereused to recognize the facial image in different ages. The SFLA was improvedto calculate the feature weights by collected facial image and savedfacial image. The weights were corrected in every recognition process,and finally the long time span invariance features were separated. Theexperiment results indicate the method based local feature and SFLA canrecognize the facial image in different ages.A method of face recognition algorithm based on local features andevolutionary algorithm was presented. The local feature matching wasconverted to an optimization problem. The different evolutionaryalgorithms were used to solve the face recogntion problem in complexconditions, it provides a new means and ideas for the research of facerecognition.