In recent years, the rise of the theory of compressed sensing signal sparsity requirements, so sparse representation of the signal received unprecedented attention. Taking into account the real signal is often nonsparsity, and compressive sensing theory must meet the requirements of the measured signal sparsity or meet at a certain sparseness orthonormal base, so sparse representation of the signal becomes very important. This paper studies explored the sparse representation of the binary mask pretreatment. The algorithm combines the binary mask the human eye is not sensitive to the removal of the DCT coefficients to improve the measurement sparsity factor in subjective image without compromising quality. Experimental results show that our proposed method reduces pretreatment CS reconstruction time and improve the quality of the reconstructed image.