Research And Implementation Of Face Recognition Algorithm Under Illumination Change And Occlusion

Face recognition has always been a research hotspot in the field of artificial intelligence image recognition,and it has been widely used in practice.For example,the train stations and airports can enter through face recognition,which speeds up the speed of entering the stations and ensures its safety.Now many electronic products have added face recognition blocks to ensure the safety of users,but the biggest problem with face recognition is that the scenes in practical applications are recognized under restricted conditions,that is to say,it is necessary to remove the glasses,hats,scarves,masks and other obstructions with appropriate lighting and the recognizers need to cooperate,and the expression cannot be too exaggerated and need to show a complete face.However,face recognition under unrestricted conditions is still the biggest problem in the field of face recognition.The unrestricted condition is not to ask for the appropriate lighting,the presence of occlusion on the face,and the large changes in expression and other factors that affect face recognition.Therefore,this paper mainly studies the face recognition algorithms based on the existence of illumination changes and occlusion under the unrestricted conditions.The main work of this paper includes two aspects:Firstly,in view of the lack of face database samples in light changes,it is difficult to express the diversity of samples.Therefore,in this paper,a virtual sample "symmetrical face" sample is added to solve the problem of insufficient samples to express the diversity of samples,such as light changes,expression changes,etc.With the increase of samples,the matching range of test samples will increase,and the recognition rate may decrease,Therefore,this paper combines the extended samples with the improved two-step face recognition(TSFR)algorithm to improve the recognition rate of face recognition with light changes.The experiments show that the algorithm proposed in Yale B,Yale A,ORL and FERET face database is superior to original TSFR algorithm based on sample expansionthe,TSFR algorithm,SRC algorithm and CRC algorithm.At the same time,in sparse representation,the advantage of dictionary learning algorithm is to reduce the running time.Therefore,in this paper,the extended samples and dictionary learning algorithm are also combined to improve the accuracy of face recognition.Experiments show that the algorithm proposed in Yale B,ORL,GT face databases are superior to some existing dictionary learning algorithms.Secondly,for face recognition with occlusion,this paper first divides the samples into five sub blocks by non-uniform partition.Then performs dictionary learning on the five sub-blocks separately,and calculates the sparse concentration index SCI of each sub-block and sorts the SCI values,assigns larger weight to the sub block with larger SCI value,and assigns smaller weight to the sub block with smaller SCI value.Finally,the five sub-blocks are combined by weighted score fusion for the final classification.Experimental results show that the AR face database with sunglasses and scarves has a significant improvement compared with some existing dictionary learning algorithms,and its recognition rate is also improved compared with CRC algorithm.At the same time,the occlusion areas are set up manually on ORL face database and illuminated Yale B face database.The experiments show that the proposed algorithm is also improved compared with other algorithms.