Face Recognition via Sparse Representation
Metadata[+] Show full item record
[ACCESS RESTRICTED TO THE UNIVERSITY OF MISSOURI AT AUTHOR'S REQUEST.] Sparse representation has been successfully used for face recognition. The sparse representation classifier (SRC) can solve illumination variations, occlusions and random noise efficiently. This algorithm mainly uses the well-known l1-norm constrained least-square reconstruction minimization technique. In this project, image pixel values are used to represent facial features. The training set is constructed as follows: each subject is presented in the training with many images representing various lighting conditions, so that a probe image (testing image) of certain illumination conditions can be represented by a sparse linear combination of the training samples. In this project, three databases are used for face recognition, including: PIE-CMU face dataset, Multi-PIE dataset and Labeled Wild Face dataset to present the advantages and disadvantages of the sparse representation classifier.