Highly overcomplete sparse coding

被引:18
作者
Olshausen, Bruno A. [1 ]
机构
[1] Univ Calif Berkeley, Redwood Ctr Theoret Neurosci, Helen Wills Neurosci Inst, Berkeley, CA 94720 USA
来源
HUMAN VISION AND ELECTRONIC IMAGING XVIII | 2013年 / 8651卷
关键词
sparse coding; natural images; overcomplete dictionaries; denoising;
D O I
10.1117/12.2013504
中图分类号
O43 [光学];
学科分类号
070207 ; 0803 ;
摘要
This paper explores sparse coding of natural images in the highly over complete regime.We show that as the over completeness ratio approaches 10x, new types of dictionary elements emerge beyond the classical Gabor function shape obtained from complete or only modestly overcomplete sparse coding. These more diverse dictionaries allow images to be approximated with lower L1 norm (for a fixed SNR), and the coefficients exhibit steeper decay. We also evaluate the learned dictionaries in a denoising task, showing that higher degrees of overcompleteness yield modest gains in peformance.
引用
收藏
页数:9
相关论文
共 16 条