Feature Extraction And Recovery Algorithm Of Face Recognition

In recent years, face recognition becomes the central issue for the portabilityand convenience of face. It is difficulty for the abstract of characteristics and nowresearchers start focus on the restoration of face image. Now a lot of algorithmsappear at home and abroad. In this paper, though the study of these algorithms,we can think about the sample’s global and local features to improve therecognition rate in the basic of sparse representation. And we combine thecost-sensitive learning and sparse representation to allow samples to identify inthe lowest risk. At the same time, we make the compressed sensing into the areaof face restoration and expand the dimensional to three-dimensional to make theeffect of recovery better.Though the study of sparse reserved photography, this paper presents globalweighted local sparse projection reserved. The method makes sample sparsereconstruction to a new sample and then projects the original sample, calculatesthe difference, obtains the sample’s between-class scatter matrix, calculateseigenvalues, obtains projection vector and project them classification.Though the study of the cost sensitive learning, the paper comes up with aalgorithm of cost sensitive sparse locality preserving projections. It pullscost-sensitive learning in the algorithm, projects samples, calculates projectiondifference, considers the cost of classifying sample incorrectly, solves theunbalance of classification, uses Laplace matrix to calculate the eigenvalues ofsamples and uses eigenvalues to classify.Though the study of the compressed sensing, this paper raises face imagerecovery based on compressed sensing using image priori knowledge to describethe nature image self-similarity of local smoothing and non-local smoothing anddiscusses the local area of piecewise smooth and non-local area of nature imageveins and structure. These keep the image local and non-local area of sparse performance and nature image consistence, and then obtain the image solutionspace.We find that the recognition rate of global weighted local sparse projectionreserved higher rate compared with typical traditional simulation, and therecognition rate of the sparse reserved photography algorithm of cost-sensitivelearning based on combining nearest neighbor classifier higher rate comparedwith traditional ways in different face library. The innovation of face imagerecovery based on compressed sensing is5db higher than others.The three kinds of simulation are recovery supplement for face identificationand recovery. They not only consider local features but also global and at thesame time they make up the unbalance of identification classification. They pullcompression in the recovery of image, and this becomes the innovation in thisarea.