Research And Design Of Facial Feature Extraction And Classification Recognition Algorithm

The face recognition algorithm has been developed rapidly in the past decades.Although the result has been fruitful,there are still a number of problems worthy of further studying.In fact,it is still difficult to achieve high recognition rate,together with stability,due to the complexity of face recognition and the susceptibility to external environment factors.In this thesis,two different feature extraction algorithms and two different classification recognition algorithms are proposed,respectively.The simulation experiment in the face database verifies the effectiveness of the new algorithms,and the maximum recognition rate is up to 99%.The main research works of the thesis are as follows:(1).Aiming at the influence of light and noise changes on face images,this thesis proposes a facial feature extraction algorithm based on Gabor filtering and second-level CS-LTP.After Gabor filtering,the image of the face is processed by CS-LTP encoding twice,then the image is extracted after the encoding,and the final classification is identified.The light experiment and noise experiment in the face database of Yale and ORL show that the new algorithm has a good robustness to the illumination change and noise suppression,and obviously improves the face recognition rate.(2).Aiming at ignored problems of global feature extraction of human face,such as partial sample outliers and the contribution to the recognition rate from important characteristic features,this thesis proposes a new algorithm combining modular PCA with improved fuzzy LDA algorithm.The new algorithm extracts feature of face images by modular PCA firstly,and gives different weights to sub-block features,then the extracted feature matrix is projected by the improved fuzzy LDA algorithm,and the final classification is identified.The simulation experiment in the Yale face database shows that this new algorithm can solve some sample outliers better,and extract feature vectors which are more conducive to classification recognition,and significantly improve the face recognition rate.(3).Aiming at the problem of low recognition rate and time-consuming for traditional classifier,this thesis proposes a new algorithm based on SVM algorithm,which is a complete binary tree SVM multi-classification recognition algorithm based on new separable measure.The new algorithm takes the new separability measure as the criterion of different classes.The SVM algorithm is applied in the complete two binary tree structure to achieve face recognition.Thesimulation experiments in ORL face database shows that the new algorithm has shorter classification time and higher recognition rate than traditional classification algorithm.(4).Aiming at the problem of the content and function of face image description for global feature and local feature,this thesis proposes a face recognition algorithm using LBP and SVM cascade global and local features.The new algorithm uses LBP encoding algorithm to encode the face image,and extracts global and local features.In the face recognition stage,the global features are first used to make a "rough" identification,and then the local characteristics are used to further identify the multiple "candidates" to obtain the identified face categories.The simulation experiment in the GT face database shows that the recognition accuracy of the new algorithm is significantly higher than a single recognition process.