Research On Robust Face Recognition Based On Regression Analysis

Face recognition technology has received much attention from academics because of its important application value,and it has become a hot topic of research in the field of artificial intelligence and pattern recognition.However,due to the complex and versatile real environment,the research of face recognition has many technical difficulties.This paper will mainly focus on the face recognition problem that is partially occluded,we find the shortcomings in the past research and propose an improved algorithm.Therefore,the main contents and innovations of this paper are as follows:(1)According to the past research on low-rank based model,this papar presents a double low-rank and self-induced representation based face recognition algorithm,which combines with the discriminate self-indeced representation classifier.Firstly,we learn the low-rank representation of the training matrix by the double low-rank representation model.Then,the discriminative self-indeced representation classifier is performed to classify the testing pictures.Finally,the experiments are expanded in the occluded AR face database and the Extended Yale B face database and the face recognition rate are calculated,respectively.Comparing with other methods,our algorithm has better validity and robustness.(2)To emphasize the importance of dictionary learning in learning representation features,this paper presents a low rank representation based dictionary learning algorithm and utilizes Schatten-p norm instead of nuclear norm in the past researches.Experimental results in two occluded databases show that the proposed algorithm can obtain the distinguishing representation features that robust to occlusion,therefore,it has better face recognition efficiency in the case of noise pollution.(3)It has been ignored in the past robust low-resolution face researches that the existence of occlusions,therefore,this paper presents a low-rank representation and locality constrained regression(LLRLCR)based face recognition method.Firstly,LLRLCR uses double low-rank representation to reveal the underlying holistic data structures for HR gallery set.For LR probe,LLRLCR uses locality-constrained matrix regression to keep regression error's structural information and to learn robust and discriminative representation features.The classification results can be predicted via a sparse representation based classifier engine.Experiments on two standard face databases have indicated that the proposed method can obtain promising recognition performance than some state-of-the-art LR face recognition approaches.