Abstract: Block compressed sensing is proposed to reduce computation quantity and storage space on image compressed sensing. There is a large influence on the whole result of reconstruction image to project and observe with the same measurement rate on every image block, because the distribution of image information is uneven, it is very different between blocks on information quantity. Based on these, a idea of measurement rate pre-allocation is proposed to apply to the image compressed sensing. The algorithm calculates every block entropy which stands for the number of information quantity in the pixel plane. Then every block will be allocated a measurement rate which matches its number of information quantity based on the power value, to change adaptively the observation number of every block. Experimental results show that the proposed method can improve the precision of reconstruction in the same measurement rate or reduce the storage space with the less observation number in the same precision of the reconstruction.

Key words: block compressed sensing, measurement rate pre-allocation, entropy estimation

摘要： 分块式压缩感知可以降低运算复杂度及运算存储空间，但是采用相同的压缩投影率进行观测投影，会影响重建图像的整体效果，因为图像所含信息量的分布是不均匀的，图像块之间所含信息量有很大的差别。基于此，提出一种投影率预分配的思想应用于图像的压缩感知。该算法通过计算每个图像块在像素平面的估计熵，来代表每个图像块所含信息量的多少，为每个图像块分配适应于本块信息量的投影率，自适应地改变每个图像块的观测值数量。实验结果表明，与分块压缩感知方法相比，在相同的压缩投影率下可以得到更好的重构质量，或者在保证重构质量的前提下，所需观测值数目更少，降低了存储空间。

关键词: 分块式压缩感知, 投影率预分配, 熵估计