In the existing adaptive multi-scale block-slice compression sensing algorithms, the role of high-frequency information in the reconstruction process is neglected, resulting in the not-complete-reconstruction of the edge contours of images. Moreover, the fixed block size is used in the process of compressing blocks, and thus the sparsity of the image itself is not fully used. In view of the above deficiencies, a multi-scale block adaptive sampling rate compression sensing algorithm is proposed. This algorithm makes full use of the high-frequency and low-frequency signals after wavelet transform, and simultaneously improves the fixed size block of images. First, the spatial filtering algorithm based on adaptive neighborhood features is used to eliminate the blockness in the low frequency part. Second, as for the high frequency part, the size of the image block is adaptively selected according to the texture features, and thus the sample block size is automatically partitioned and the sampling rate is adaptive. Finally, the images with different amounts of texture information are compressed and reconstructed. The results show that the reconstruction effect by the proposed method is obviously superior to those by the existing adaptive sampling rate algorithms.