Research On Partially Occluded Face Recognition Based On Single Sample

The existing facial recognition methods have achieved satisfactory performance under well-controlled condition.However,face recognition is widely used in business and law enforcement fields.In these applications,it is necessary to recognize face images obtained from surveillance cameras and handheld devices.In such real scene,it is difficult to require user's cooperation.So,the captured face images may contain occluded faces,which makes face recognition difficult.And in real life,due to the difficulties in collecting image samples or the storage limitations of the application system,each person may have only one face image in the train set.This problem is called single sample problem.In the case of the single sample database,many currently partially occluded face recognition methods can not perform well,and the single sample partially occluded face recognition is common in practical applications.Therefore,in order to deal with single sample partially occluded face recognition,the single sample partially occluded face recognition method based on keypoint cluster blocks weighting strategy is proposed in this thesis.The main work of this thesis is as follows:1)First,keypoint detection and matching,density-based clustering,texture feature,and image entropy are introduced in this thesis,which provide a theoretical basis for the method proposed in this thesis.2)Secondly,the single sample partially occluded face recognition method based on keypoint cluster blocks weighting strategy is proposed in this thesis by using several methods such as keypoint detection and matching methods,density-based clustering method,texture feature and image entropy.The basic flow of the method is as follows: give a test image and a training image,and SIFT method and SURF method are firstly used to obtain the SIFT matched keypoints and SURF matched keypoints between the image pair.Secondly,a keypoint cluster block acquisition method is proposed to merge the neighborhood of matched keypoints in each image.And,image block sets of the test image and image block sets of the training image are obtained.Thirdly,a weighted descriptor matching strategy is designed to obtain matched block pairs of two image block sets.Finally,the average distance of all matched block pairs is calculated,and the average distance is re-weighted with the number of matched keypoint pairs to obtain the similarity of the face image pair.The experiments prove the effectiveness of the proposed algorithm for single-sample partially occluded face recognition.3)Finally,the face recognition defense system developed,which is robust to partial occlusion,is introduced in this thesis.The system is designed and developed based on the proposed method.First,UML was used to analyze and design the face recognition defense system.Then,the system developed based on UML-based analysis and design is introduced.Finally,the system performance was tested.The test results show the practicability and accuracy of the developed system.