Research On The Key Technology Of Face Recognition Person-witness Matching Based On Deep Learning

With the development of science and technology,the use of person-ID matching technology is becoming more and more frequent,and the face recognition technology based on convolutional neural network has shown great advantages over traditional algorithms in relevant applications,and it is the focus of this paper to push the face detection and face recognition algorithms to further improve the performance and efficiency of person-ID matching.In this paper,2 solution ideas are targeted,and the main research content includes the following three parts:(1)To address the problem that the original image noise affects the detection accuracy of the model,this paper proposes a face detection algorithm WT-IMTCNN by combining the characteristics of wavelet transform that can denoise and MTCNN that learns image features autonomously.Since the fixed threshold used in the original wavelet change is unreasonable in denoising and affects the model performance,this paper proposes a hybrid threshold function and adaptive thresholding to improve the wavelet transform to achieve image denoising without distortion;in addition,since the MTCNN network uses a non-maximum suppression screening algorithm for candidate frame screening will lose some information,to address this problem,this paper proposes a confidence level to improve the NMS algorithm,which is applied to MTCNN to form the IMTCNN detection algorithm.(2)To address the problem of poor face recognition accuracy,this paper proposes a face recognition algorithm GI-FR.In order to enhance the potential information and expression ability of the original image,this paper calculates the 8 feature compensation factors of the original image by 8 common operators,and the generated 8 feature description images are weighted and fused to achieve image information enhancement.For the optimal combination of feature compensation coefficients of the 8 feature compensation factors,this paper uses the GA algorithm to achieve the selection of the optimal combination,and the new feature map is dimensionally reduced and input to the traditional classifier SVM to achieve high accuracy face recognition.(3)Simulation experiments are conducted using Wider Face dataset and LFW dataset to verify the detection performance of the MT-IMTCNN model and GI-FR model proposed in this paper,the results showed that the maximum detection rate could reach 98.8% and99.3%,which is an increase of 9.9% and 0.2%,respectively,compared with existing studies.and multiple sets of comparison experiments are reasonably set to verify the correctness of the adaptive thresholding,improved NMS algorithm,image feature compensation techniques proposed in this paper,and the application effect on multiple datasets,and the experimental results are analyzed and organized in detail.