In image processing, dictionary learning provides good high performance on the image classification problem. Inspired by dictionary learning of sparse coding, we propose a new algorithm to learn the sparse representation and corresponding dictionary. Firstly, we modify Graph Regularized Sparse Coding (GSC) to learn dictionary atom with more coherence within classes. Then we learn class-specific dictionaries and apply principal component analysis to these dictionaries. Finally, we can obtain new feature representation by projecting these class-specific dictionaries into sub space. The experimental results show that our method presents significant performance improvement in comparison with many existing methods.
為了持續優化網站功能與使用者體驗,本網站將Cookies分析技術用於網站營運、分析和個人化服務之目的。
若您繼續瀏覽本網站,即表示您同意本網站使用Cookies。