The Research Of Locality Preserving Projection Based On Sparse Representation

In recent years, face recognition technology obtained the rapid development and greatprogress, some better recognition methods have been proposed by researchers, this largelypromotes the development of the application of face recognition technology, the current facerecognition technology has entered the stage of practical application. Face recognitiontechnology has a wide range of applications, it is widely used in the identification, publicsecurity case investigation, entrance guard system, attendance system, games, entertainmentand other fields, it also makes the human face recognition technology become a hot researchtopic today.The process of face recognition is not very complicated, can be generally divided intoseveral steps: face detection, feature extraction, classification matching. Face detection isdetect face image from video sequences or static image, in preparation for the next step of thefeature extraction. Generally, faces is a relatively high dimensional data, for example, theimage pixels are used in the process of dealing with an analysis of face images, for a smallface image with165×120dimensions, need to save the pixel value is19800, this dimensionis large, and now the picture resolution is far greater than this dimension. So, in view of thiskind of high-dimensional data, in order to reduce the amount of calculation and the time, thehigh-dimensional data need dimension reduction, is also the process of feature extraction, thisis one of the most important steps in the process of face recognition. Feature extraction is theprocess of dimension reduction, it is a transformation form high dimensional data to lowdimensional data, the process can not only to reduce the amount of calculation for theclassification of matching, and can remove some redundancy information of thehigh-dimensional data, which is not have the ability to distinguish. The extracted features canbe matched through the classifier, to obtain recognition results.This article is based on the research of the locality preserving projection (LPP) algorithmand sparse representation method, and then combining these two methods for face recognition.For high dimensional face data, using locality preserving projection (LPP) algorithm toreduce the data dimension, this method can keep the local neighborhood relationship consistently between samples before and after the dimension reduction, and effectivelyguarantee the relationship invariably for the high-dimensional data in projection to lowdimensional space. After the data dimension reduction, we use sparse representation methodto classify the feature data, achieve the result of recognition. Due to sparse representation hasstrong robustness to the part of occlusion for the face image and uneven illumination, so usingthe sparse representation is used to obtain a better identification effect than the originallocality preserving projection (LPP) algorithm for classification. By combining the twomethods, this new method combines the advantages of the two methods, combines thecharacteristics of the two methods, and effectively improved the recognition rate of thealgorithm. By experimental analysis on several recognized face database: ORL face database,Yale face database and the AR face database, we arranged several group experiments on theoriginal locality preserving projection (LPP) algorithm and our new method, through theexperiment results, it shows that our proposed method relatively obvious enhancement thanthe original locality preserving projection (LPP) algorithm on recognition rate, and obtain agood recognition effect.