Research On Robust Learning Based Face Recognition Technology

As the big data and the artificial intelligence develop continuously,multimedia big data is increasingly close to our life and the society.Among them,face recognition with its information collection distance,easy to be stolen,face payment,face unlocking and other important academic and industrial fields of unique advantages of attention.However,in the real-world environment,the face image is going to be influenced by many elements like occlusion,expression as well as illumination,which influences face recognition's performance to a great extent,causing a severe decline in face recognition rate.This paper is going to pay attention to how to promote the recognition rate when the face is blocked,detailed studies and discover the disadvantages of the algorithm before in order to make improvements.The core innovations and contents of the paper are shown below.(1)The paper explores and develops the previous image blocked-based classification models,and proposes a set of face recognition methods based on multi-scale features to represent the face images contained in different parts of the context semantic information of the learning combination of characteristics.Firstly,the face image can be divided into some image blocks.Later on,such image blocks can be robustly represented.Finally,considering the effect of image block size on recognition rate,multi-scale fusion algorithm is applied to our method.Experiments on LFW,AR and extended Extended Yale B face database have completely demonstrated such strategy's recognition rate.By comparing with traditional strategies,the suggested strategy has nice efficiency and robustness.(2)Considering the noise,the context similarity of noisy image blocks can usually provide some auxiliary information for recognition.A robust face recognition method is proposed based on the multi-scale manifold structure preservation.This paper makes full use of nonlinear structural similarity of face images in manifold space.The method assumes that the image block in the center of the window has the same manifold structure as that in the context in the segmented image window.Based on this assumption,the central image block is connected with its context image block to obtain a new set of image blocks,and then the representation learning is carried out at the same position of the training sample.Finally,the identification results of different scales are fused.Experiments on LFW,AR and extended Extended Yale B face database have completely demonstrated the suggested strategy's robustness and efficiency.(3)For the purpose of making up for the drawbacks of conventional artificial feature extraction algorithms and fully using deep learning face recognition technology's strong points,this paper improves the conventional algorithms and proposes a face recognition method based on local adaptive convolution feature joint collaborative representation.The model makes full use of the local information of the face image,extracts the local characteristics of each image block by using trained lightweight convolutional neural network training,and then learns the local adaptive convolutional features of all image blocks to obtain the test sample of the representation coefficients of each local area of the test sample together,for the purpose of completely recognition and classification.Experimental outcomes on LFW and AR face databases present the suggested strategy's stability and strong points.