Research On Face Recognition Algorithm Based On Group And Collaborative Methods

With the development of science and technology,biological feature has been widely leveraged in pattern classification problems and face feature is a category of crucial biological feature and is being studied not only in industrial community but also in academic community.A large number of individuals aim to make the face recognition algorithms or products more stable and reliable.Unfortunately,reliable face recognition algorithms or products are relatively rare.As we all know,due to the environmental factors,such as light,angle and posture,and various face characteristic such as hair,expression and age,face recognition is still confronted with great challenges.In the departed decades,there were massive famous algorithms which have been proposed and their performances are excellent.Some of them are:Turk et al.have proposed principal component analysis(PCA)face recognition algorithm.Fisher also proposed an algorithm named Fisher's discriminant analysis(FDA)face recognition algorithm.Yang et al.proposed two-dimensional principal component analysis(2DPCA)face recognition algorithm.In the beginning stage of face recognition,these algorithms played crucial roles for the development of face recognition algorithms.Furthermore,even in now,the methodology of these algorithms still shows great research value.With the development of computer science,people have known that these algorithms are hard to meet their demands such as face detection in access control system and face recognition in public security system.It is clear that these applications are suitable for plain scenes.Recently,compressed sensing(CS)theory is sufficient expanded.The sparsity of image and video has been mined out.Face recognition using sparsity technique is becoming rapidly popular such as sparse principal component analysis(SPCA),robust face recognition via sparse representation(SRC).However,information production and spread are becoming increasingly diverse due to the intelligent terminals.Folks are surrounded by big data.Then,how to effectively harness the information is urgent to solve especially in the field of face recognition.Based on this idea,we do the following work:(1)This paper elaborately and detailed summarizes the research background and the significance of face recognition at home and abroad.Then this paper narrates the merits and challenges of current classical algorithms.(2)In view of compressed sensing(CS)theory,we propose a joint sparse coding(JSC)algorithm.We find that the accuracy will greatly increase when the sparse algorithm utilizes multi-situational sample.The experimental results demonstrate that our JSC algorithm is robust than traditional sparse algorithms.(3)It is clear that sparse algorithm is time-consuming.Based on projective dictionary pair learning(DPL),we employ the group idea and collaborative thought and propose group and projective dictionary pair learning(GDPL),collaborative and projective dictionary pair learning(CDPL).(4)By summarizing GDPL and CDPL,we find that group method and collaborative technique can be facilitated together,so we further propose an algorithm named group and collaborative dictionary pair learning(GCDPL).AR face dataset,Extended Yale B face dataset are widely used in face recognition simulation.Isolet speech dataset is also widely utilized in pattern recognition.This paper also carry out elaborative experiments based on these datasets because these datasets are universal accepted not only in industrial community but in academic community.The final experimental results also demonstrate that our proposed algorithms are robust and effective.Related researches have been accepted by ICDE international conference and ICPR international conference.