Face Recognition Based On Regression With Norm Regularization

Face Recognition (FR) has been a concerned issue because of its wide application in information science field in recent years and new technology about FR also emerge one after another. However, novel robust and effective algorithms about FR have important practical meaning facing with complicated practical environment and high precision demand.Regression algorithms based on norm regularization have been applied widely in machine learning and pattern recognition. The norm constraint can minimize the error function and lead the solution towards the least gradient direction. There are several algorithms based on regression with norm regularization including SRC, LASRC, RSC, CRC_RLS, NRS, SPP, CRP that have good performances. In this paper, we study the regression algorithms based on l1,l2,l21 and l? norm regularization, which are used in face classification and feature extraction. LASSO regression based on l1 norm can extract adaptively sparse feature of data, hence SRC is robust on to occlusion. While a probe image belongs to a pattern eventually, so sparsity intentionally is more meaningful than normal sparsity. The paper proposes class sparsity with references to group sparsity and l? norm.To overcome the problem of multicollinearity in LRC which also can be regard as a NS classifier, the paper introduces NRS to the field of FR. Meanwhile, NRS-LDA classifier is proposed by using LDA to construct Tikhonov matrix. We reveal the advantages and disadvantages of NS, CRC-Pre, NRS and NRS-LDA with detailed experiments and analysis. Besides, A adaptive nearest regularized subspace (ANRS) is proposed by combining LRC with NRS.To improve the robustness of LRC and PCR to strong illumination, we put forward 2DPCR algorithm. Firstly, all face images are partitioned into several blocks and the approach performs 2DPCA process to project the blocks onto the face spaces. Then, LR is used to obtain the residuals of every block by representing a test image as a linear combination of class-specic galleries. Lastly, three minimum residuals of every block and fuzzy similarity preferred ratio decision method are applied to make a classification. The proposed framework outperforms the state-of-the-art methods and demonstrates strong robustness under various illumination, pose and occlusion conditions on several face databases.SPP holds the sparse relationship between data, and projections contains natural discriminating information. Unfortunately, Calculating the l1 norm is too complicated, so CRP use l2 norm replace l1 norm to improve computational speed. Compared with SPP and CRP, LRR is better at capturing global structures of data and good at handling corrupted data. Therefore, to take the place of SR and CR with LRR, discriminant Low-Rank presentation projection (DLRRP) is proposed. Meanwhile, bound term of inner class distance is also introduced to enhance discriminant ability.