Research On Face Recognition Algorithm Based On Unrestricted Case

As one of the important components of computer vision,face recognition plays a crucial role in public safety.In the process of face detection,the accuracy of face recognition algorithm decreases due to the constraints of human interference and environmental factors.Therefore,the main work carried out in this paper,with face recognition as the research,is as follows:Aiming at the situation that face recognition accuracy is not high and efficiency is low in face recognition under non-limited conditions such as sudden change of light,uneven image quality and occlusion.A Siemase neural network based on Local Binary Pattern(also called LBP)and Frequency Feature Perception(SN-LF)model is designed.This network is based on Siemase networks and uses the Uniform LBP algorithm and Frequency Feature Perception to achieve face recognition under unrestricted conditions.Algorithm LBP eliminates the effect of illumination on the image while providing vector-level input to the network model;Frequency Feature Perception classifies image features into low-frequency features and high-frequency features.The low-frequency features in the Siemase neural network are compressed to increase the recognition efficiency of the network,while information is exchanged with the highfrequency features,so that the target noise data is eliminated while retaining its feature data.That maintains the recognition rate of the network and improves the computing speed of the network.Simulation experiments are conducted on standard face datasets Yale-B,CASIAWeb Face and LFW,and compared with other network models.The experimental results show that the proposed SN-LF network structure in this paper can improve the recognition accuracy of the algorithm and obtain a better recognition accuracy.To address the problem that occlusion and other situations occur in the face recognition process leading to the algorithm’s wrong and missed detection,a face detection algorithm based on spatial pyramid pooling and fused with dictionary learning is designed in this paper.In the data pre-processing stage,data enhancement is performed on the image,and the random erasure algorithm is used to randomly erase the target in the image to increase the target occlusion data set,then deep-level feature data is extracted from the target and the occlusion target respectively by spatial pyramid pooling,and finally the feature data is represented sparsely and the complete dictionary is obtained by fusing the occlusion face feature dictionary and the non-occlusion face feature dictionary to improve the target in the target in case of interference such as occlusion.Simulation experiments are conducted on the standard face datasets Yale-B and LFW datasets and compared with other face recognition algorithms,and the experimental results show that this paper still has a high recognition rate when the target is obscured.