Research On Face Feature Extraction And Recognition Application Based On Ensemble Learning

With the development of artificial intelligence and scientific information technology,and the needs of national security and people’s life,the face recognition technology has received extensive attention and has achieved great development.It is one of the research hotspots in the field of pattern recognition and computer vision.Because of the influence of face pose,illumination,expression,age and small sample,face recognition is very challenging,and can not meet the needs in many cases.Continuous research is also needed on How to extract robust and highly classified face features and recognition.This paper mainly focuses on the research of face feature extraction and recognition on the small sample problem by using the characteristics of integrated learning,which enhance the generalization ability of recognition algorithm,to enhance the recognition performance.The specific research work is as follows:(1)A subspace face recognition method integrating improved Speeded-Up Robust Features is proposed.Due to the problems of SURF algorithm like mismatching in feature matching,we use the shape model of AAM to improve and obtain the local features of the face image.Then,according to the idea of combining different learning models in ensemble learning,the extracted local features are integrated with the global PCA features to obtain the facial features which include global and local information and will be used for matching recognition.Finally,experiments were performed on ORL face database and FERET face database to verify its recognition performance.Compared with other classical algorithms,the recognition rates of proposed method were improved to different degrees.(2)A multi-featured integrated face recognition method is realized.In order to extract and describe the features of human faces more comprehensively,the globalPCA features,local LBP features,and GIST features that can obtain image context abstracts are extracted from the face images respectively.According to the idea that ensemble learning can improve the generalization ability of the algorithm by changing the sample set,the image is processed by Gauss down sampling,and three features are extracted at each scale,and then the majority voting method is used for integration.Finally,experiments on ORL and FERET face database show that the recognition performance of the realized method is improved.(3)The convolution neural network model for deep learning can learn face features independently without human design in face recognition,but it requires training of a large number of samples,and the recognition performance of the face recognition is seriously reduced in the case of small samples.A face recognition method with integrated deep learning is proposed.First,face images are scale transformed to form ten different scale images.Then,traditional convolutional neural networks are trained at each scale.Nextly according to the Stacking thought of ensemble learning,a BP neural network is trained by using the output probability value of each network’s output layer as a meta-feature,thus the final recognition results are obtained.Finally,this method is tested on ORL face database.The results show that this method has better performance than traditional convolutional neural network in the case of small samples.