Low Rank Character Description And Image Classification Method Research

With the continuous development of information digitization,the large data become a hot topic in daily life,image occupies a large proportion in the large data,therefore,image classification technology has entered a rapid development period.as a classic question,Image classification is still a research subject by many scientific researchers,and has been widely used in pattern recognition such as network technology,information security and public security area.Existing image classification algorithms,however,in the case of the training sample pollutted by noise pollution,the image recognition rate is reduced,so that the classification of the algorithm performance.To solve these problems,this article main research content is as follows:(1)In view of the training samples pollutted by noise pollution,PCA algorithm describe the high dimensional samples with a small number of the characteristics,effectively obtain the most important element and structure from the "rich" data information,and remove the noise;By using unit matrix as error dictionary,SRC algorithm can effectively describe or present the sample image which exists error or damaged;By using a complete dictionary as error dictionary,RPCA algorithm calculate the sparse representation of the sample data under the dictionary.Through image classification performance contrast of the three basic algorithm,highlight the advantages of the RPCA algorithm,and show that the low rank description based on image features can solve the problem of image affected by noise pollution.(2)The low rank feature of LRR algorithm is very similar with the representation feature of RPCA algorithm,both are described based on the image low rank characteristic,studies have shown that LRR algorithm is more effective and more accurate to extract low rank features of the image.Because the data often interference by noise,need to extract more child structures of the data,but RPCA algorithm describe all the characteristics of samples date by one child space,so such data description is not accurate.So,low rank description(LRR)not only can extracted the common features of the sample data from different subspaces,also eliminate unstable characteristics.(3)Under the condition of the sample images interference by large noise,LRR algorithm is very sensitive to local noise,which significantly reduced image classification performance,therefore,this paper proposes an improved low rank algorithm,this algorithm can effectively remove noise or error from sample data,which make image restoration more accurate,also improve the rate of classification recognition.Differs from the traditional low rank algorithm this algorithm is introduce a constraint in the original algorithm model to ensure the sparse of the coefficients,obtain the geometric structure of the image data in each space,its effective solution is through the augmented Lagrangian multiplier method(ALM).Especially under the condition of sample image interference by large noise,this algorithm compared with other algorithms can keep higher recognition rate,and has a great advantage.Finally,this paper will apply the improved low rank algorithm to the face image,the algorithm extract inherent characteristic information of the image then classification,and simulation on the ORL and Yale B face database is carried out to verify the algorithm,the results show that this algorithm has strong anti-interference ability,which improve the recognition rate.