Face Recognition is considered as an important technology of modern biologicalinformation recognition. The traditional face recognition method is easily affected bynoise, shade, and so on. The sparse representation based classification for face imageshas been one of efficient approaches for face recognition. And the method is morerobust to noise and occluder compared with the traditional algorithm of face recognition.However, existing sparse representation based classification for face images ignores theintrinsic geometric property of sample in dictionary learning, which results in theproblem that the coefficient values corresponding different categories atoms are large,and the recognition is not good enough. In addiction, dictionary learning ignorescoefficient distribution characteristics, which will impair the recognition performance.Paper’s main work is as follows:â‘ Dictionary learning based on classified K-SVD ignores the similar geometricproperty of samples, which leads to the problem that the coefficient valuescorresponding different categories atoms are large, and the recognition is not goodenough. Hence, learning a discriminative dictionary via fusing the similarity of thesamples based on classified K-SVD(FSS-KSVD) is proposed for this problem. Welearn a set of atoms for each class of face images. Then all the classes of atoms canbe concatenated into one dictionary. At last, FSS-KSVD is applied to facerecognition. The experimental results show that the proposed algorithm is betterthan K-SVD and classic SRC algorithm, and it achieves good recognition effect.â‘¡FSS-KSVD algorithm ignores the coefficient distribution characteristics anddiscrimination of the atoms of different category, which may result in the problemthat the reconstruction error on the atoms of in the same category is larger, and theperformance is not good enough. Hence, learning a supervised and discriminativedictionary for face recognition(LSDD) is proposed for this problem. We learn adiscrimination dictionary with class label information. At last, LSDD is applied toface recognition. The experimental results show that the proposed method issuperior to the SRCã€KSVDã€DKSVDã€LSRC〠FDDL and LSDL in the ARdatabaseã€Extended Yale b database and ORL database. |