| Summary: | Sparsity-based algorithms recently have received great interests from statistics, signal processing, machine learning as well as computer vision. In this master thesis, it discusses the sparse representation based algorithms for computer vision problem, including the independent sparse representation (ISR), locality-constraint coding, group sparse representation (GSR). Based on these existing algorithms, two new algorithms referred to as locality-constrain group sparse representation (LGSR) and multiple-kernel group sparse representation (MKGSR) are proposed. Comprehensive experiments for Human Gait Recognition (HGR) using USF HumanID Gait database show that the two newly proposed methods, LGSR and MKGSR respectively achieve the best Rank-1 and Rank-5 recognition accuracy.
|
|---|