Human Ear Recognition Algorithm Based On Local Feature And SRC Fusion

Traditional identification mainly uses external objects,which are easy to be stolen or lost,such as passwords,certificates and keys for identity verification.Therefore,the advantages of biometrics gradually begin to appear.The uniqueness,eternity,security and convenience of biometrics have attracted wide attention of researchers in recent years.Ear recognition is one of the most important recognition ways in biometrics.As an emerging biometric identification,human ear recognition has the following advantages: ear shape hardly changes with age;ear shape does not change much when facial expression changes dramatically;ear and face data can be obtained together,laying a certain foundation for multimodal recognition.In addition,ear recognition is also widely used in criminal tracking and forensic verification.It is in this case that we propose an effective ear recognition scheme based on local feature-based feature-level fusion and sparse representation-based decisionlevel fusion.The main work is as follows:1.Human ear recognition is introduced from four aspects: human ear extraction,preprocessing,feature extraction,and classification and decision.Development trends and main challenges of ear recognition are also explained in detail.2.Preprocessing,feature extraction and classification technologies based on unimodal ear recognition are introduced,and the effectiveness of each preprocessing,feature extraction and classification technology in the field of human ear recognition is verified based on relevant experimental results.In addition,we also introduces the categories,fusion rules,and the latest research of multimodal recognition based on inter-level and intra-level fusion.3.An effective ear recognition algorithm is proposed,including preprocessing,feature extraction,and classification and decision.The algorithm is tested on 6 data sets of USTB1,USTB2,USTB3,IITD1,AMI and AWE,and obtain the accuracy of 99.44%,97.08%,100%,100%,98.14% and 82.90% respectively.Compared with the related and latest algorithms,experimental results prove that proposed algorithm has certain advantages.4.The challenges based on occlusion and pose changes are analyzed,the robustness of proposed algorithm to occlusion and pose changes is tested,and the solution for multi-pattern recognition based on inter-level fusion is proposed.Experimental results verify the feasibility of multi-pattern recognition.