摘要
自适应压缩感知(adaptive block CS,ABCS)利用方差表征图像纹理复杂度,进而对图像块进行分类并分配不同测量率,能够在总测量率一定的情况下较好重构图像。但该方法并没有考虑到图像块中的结构特性。针对这一问题,本文提出了基于图像块分类及自适应多字典学习的图像压缩感知方法,先将图像块分为平滑类、边缘类及不规则类三类,再根据结构特性对边缘类进行进一步细分类,针对各类图像块分别训练各自对应的自适应字典。实验结果表明,与单字典和仅基于方差分类的多字典算法相比,本文设计的方法明显提高了图像的重构质量。
Adaptive block CS used variance to represent image texture complexity, then classifies image blocks and assigns differentmeasurement rates, and can get a better reconstructed image under certain measurement rate. However, the image structure is ig-nored in this method. According to this limitation, this paper proposed a new compressive sensing method based on image blocksclassification and adaptive multi-dictionary learning. First, image blocks are classified as smooth blocks, edge blocks and irregularblocks. Then, the edge blocks are further classified according to the structural characteristics. Finally, to train corresponding adap-tive dictionaries for each kind of blocks. Experimental results show that the proposed method significantly improves the reconstruc-tion quality compared with single dictionary method and adaptive block CS.
出处
《电脑知识与技术》
2018年第2Z期190-193,共4页
Computer Knowledge and Technology
关键词
压缩感知
块分类
分类字典学习
compression perception
block classification
classification dictionary learning