Research And Application Of Single Sample Face Recognition Algorithm Based On Multi-layer Voting

Single sample face recognition has a broad application prospect in people's lives,such as law enforcement and dynamic monitoring and access control.The single sample per person(SSPP)problem is a great challenge for real-world face recognition systems,because there is always a large gap between a normal sample enrolled in the gallery set and the non-ideal probe sample.In the traditional face recognition algorithms,the improvement of recognition performance mainly depends on the expansion of the gallery set,however,in the most real-world situations there is only one image per person available.Moreover,with the linear expansion of the gallery set,the time cost of the face recognition algorithm tends to increase nonlinearly,which greatly limits the application scenario of these algorithms.Therefore,it is of great significance to study the face recognition algorithm under single sample condition.This thesis designs and implements a face recognition attendance management system for employee attendance management.In a real face recognition application system,the face recognition module is the most core one.However,besides the face recognition module,we also require an auxiliary motion target detection module to detect and segment the moving target(or face)from an image before we perform face recognition.In this thesis,we study the face recognition module and the moving target detection module,and propose new solutions.The main work of this thesis is described as follows:This thesis proposes a multi-level voting based single sample face recognition algorithm(MLV).Most single sample per person(SSPP)problems can be considered as classification ones in terms of machine learning,which can predict the category of a given sample according to the existing training samples.The proposed method is more in line with human cognitive law.MLV can avoid wasting too much time in the training process as traditional single sample face recognition algorithms,and solve the problem of incremental learning which means it has a low expansion cost when adding sample images of new categories to the gallery set.Experimental results show that the proposed MLV algorithm has the advantages of high recognition rate,fast training and low expansion cost.This thesis proposes a block-based background modeling(3BM)algorithm for moving target detection using frame difference and background subtraction.First,we divide each image into multiple non-overlapping local blocks and construct a reliable background image according to the frame difference of each small block.Then we use the background subtraction method to detect moving targets.Through extensive experiments,we find that our proposed method has the following advantages: stability,fast detection and high accuracy of moving target detection.This thesis designs and implements a face recognition attendance management system based on the proposed MLV and 3BM algorithms.This system not only realizes the face attendance function but also creatively puts forward a moving target detection-based human-computer interaction system,which is convenient for the employee to use our system and can reduce the dependence of face recognition module on hardware.Through our system,the employee can be more convenient to complete the attendance operation,and the administrator can also be more convenient to complete the attendance management.