Use Matlab

Files is in this link ORL.mat https://drive.google.com/file/d/0B_W5TGttGNmzYmRYUThaVjhyS0E/view

Sparse representation is very useful in face recognition. Here we implement based on this paper: Robust Face Recognition via Sparse Representation". There are two basic steps in this approach: 1) learning a sparse representation for each test face 2) classification based on the sparse vectors, test face, and training dat dXn. is a matrix including all training data, d is the dimensionality of face feature Assume A E R and n is the total number of faces. Given a test sample y, we would like to find the sparse vector B satisfying: minll ylli 1llBlli The learned nx1 sparse vector B indicates the importance of each training sample in reconstructing the test face. To find the identity for test face y, we use the followed formula: minllAiBi -yll3, where "i" is the identity label, and Ai, Bi is the subset of A,B corresponding to identity "i" In this question, we use ORL face dataset for evaluation, which includes 40 people and 400 images. The size of each face is 64x64. For each person, the first 5 images will be the training images, while the rest 5 will be the test images. For better performance, you may consider using PCA or LDA as a pre-processing step before running sparse representation Please report the recognition accuracies for: (1) raw pixels (64x64 face (2) pre-processing by PCA, (3) pre-processing by LDA. [provided files list]: ORL.mat Sparse representation is very useful in face recognition. Here we implement based on this paper: Robust Face Recognition via Sparse Representation". There are two basic steps in this approach: 1) learning a sparse representation for each test face 2) classification based on the sparse vectors, test face, and training dat dXn. is a matrix including all training data, d is the dimensionality of face feature Assume A E R and n is the total number of faces. Given a test sample y, we would like to find the sparse vector B satisfying: minll ylli 1llBlli The learned nx1 sparse vector B indicates the importance of each training sample in reconstructing the test face. To find the identity for test face y, we use the followed formula: minllAiBi -yll3, where "i" is the identity label, and Ai, Bi is the subset of A,B corresponding to identity "i" In this question, we use ORL face dataset for evaluation, which includes 40 people and 400 images. The size of each face is 64x64. For each person, the first 5 images will be the training images, while the rest 5 will be the test images. For better performance, you may consider using PCA or LDA as a pre-processing step before running sparse representation Please report the recognition accuracies for: (1) raw pixels (64x64 face (2) pre-processing by PCA, (3) pre-processing by LDA. [provided files list]: ORL.mat