摘要
为了提高图像稀疏表示性能,提出了一种有效的结构化字典图像稀疏表示方法.针对过完备字典构造和稀疏分解中原子筛选问题,提出了一种基于灰色关联度的字典原子筛选和结构聚类方案.首先,对测试图像分块处理,利用块作为原子样本;然后,计算原子间的灰色关联度,并设置原子灰色关联度的筛选准则;最后,利用结构特征对原子聚类,构造图像稀疏字典.算法利用灰色关联度选择表征能力强的原子,提高字典的表征能力,缓解了传统字典设计对原子个数的依赖;同时,降低了算法的复杂度.将该方法得到的字典用于图像去噪,结果表明,视觉效果明显优于同类算法,峰值信噪比提高2 dB左右,且算法复杂度显著降低.
In order to improve the performance of image sparse representation, an effective represen- tation method based on a structured image sparse dictionary is proposed. For the atomic selecting problem in constructing an over-complete dictionary and sparse decomposition, a kind of images dic- tionary atom selecting and structure clustering method based on the grey relation is proposed. First, atoms are selected by the blocking test image. Then the grey relation between atoms is calculated, and the rule of atom selection is set up. Finally, atoms are clustered by characteristics of structures, and a sparse representation dictionary is constructed. In this method, atoms with high representative performance are selected by the grey relation, improving the ability of capturing the image structure. The method solves the problem of a traditional dictionary designed depending on the number of se- lected atoms and also reduces the complexity. The simulation results show that the proposed method is superior to other algorithms in objective quality. The peak signal to noise ratio value increases about 2 dB. And the complexity of the algorithm is decreases greatly.
出处
《东南大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2014年第1期116-122,共7页
Journal of Southeast University：Natural Science Edition
基金
国家自然科学基金资助项目(61171077)
江苏省高校自然科学研究资助项目(12KJB510025
12KJB510026)
南通市引进人才资助项目(03080415)
南通大学创新人才基金资助项目(2009)
关键词
图像稀疏表示
原子优化
灰色关联度
原子聚类
image sparse representation
atomic optimization
grey relation
atomic cluster
