The salient texture structure of actual images can provide priori information for Block Compressed Sensing (BCS) algorithm and optimize the algorithm. Based on this, a new Adaptive Block Compressed Sensing (ABCS) algorithm based on saliency is proposed. The saliency in the proposed algorithm is built on the theory of gray-level spatial-dependence matrix and Weber's theorem. The deterministic Orthogonal Symmetric Toeplitz Matrix (OSTM) is adopted to measure the target image. The adaptive block strategy minimizing the average entropy, the block-vector generation method maximizing the angle second-order moment and the adaptive sampling rate setting under the synthetic feature are proposed. The analysis and verification are carried out by using different basic reconstruction algorithms. Experiment results show that, compared with the traditional algorithm, the proposed algorithm performs better on different indexes, is easy to implement by hardware and has universality and stability for different reconstruction algorithms and test images.