Face Recognition Technology Based On Integrated Neural Network And Multi-information Fusion

The technique of face recognition is one of the important intellectual recognitiontechniques which are applied to human identification in the fields of video monitoring system,finance as well as security and protection. The information security of modern society hastensthe fast development of face recognition technology. At present, the researches of facerecognition algorithms have obtained better results, however, due to the complex environmentduring the shoot, lighting conditions, the human face with glasses and beard and other obstaclesas well as expression and posture, which will influence the speed, accuracy and robustness offace recognition algorithms can not fully meet the needs of people in practice.Face recognition which select the input face image data and extract important features forclassification and identification is a technology. At present, the main research direction of facerecognition: face data preprocessing, feature extraction, classification and recognition. Inrecent years, neural network ensemble learning has become a hot research in the field of neuralnetworks. Neural network has a prominent ability to solve regression and classificationproblems, and the integration of neural network will be better generalization ability than asingle neural network through training several neural networks. Therefore the neural networkwas applied to face recognition in this paper. Face feature extraction is a key step in facerecognition. Feature extraction and classification were analyzed and studied deeply in thispaper. The algorithms of kernel independent component analysis (KICA) and kernel principalcomponents analysis (KPCA) were respectively used to extract features, and KICA algorithmwas also improved. At last, we chose neural network as classifier to fuse the features extractedby KPCA and KICA algorithms. The classification results were outputted by relative majorityvoting.KICA algorithm which resolved the shortcomings of the ICA algorithm in the nonlinearextraction and also has better robustness of the ICA algorithm in facial feature extraction is anon-linear approach. In the paper, we combined the fifth-order convergence of Newton iterationmethod and the steepest descent method to reduce the dependence on the initial matrix oforiginal algorithm and we also added a penalty factor in the algorithm, which applied to the coreof the iterative process of the FAST ICA algorithm. So an improved KICA algorithm was obtained.We extracted representative local features from face data by using the improved KICAalgorithm. KICA algorithm is a combination of ICA and KPCA algorithm, so the time of facialfeatures computation was saved and the amount of computation was reduced when we usedKPCA algorithm to extract the global features. At last, we integrated BP and PNN neuralnetwork to fuse the features extracted by KPCA and KICA algorithms. The results wereoutputted by relative majority voting. The individual learning differences are lager and theintegrated effects are better, because of the facial features extracted by KICA and KPCAalgorithms. Experiment on the public face database, the result showed that the algorithmproposed in the paper has strong robustness, higher efficiency of the implementation andrecognition rate.