Based on the compressed sensing theory, the MP (Matching Pursuit, MP) algorithm was applied to image compression coding. This paper presented the reconstructed method of the atomic library and reduced the computational complexity of the traditional MP decomposition by using the decomposition method of the block compressed sensing image. Aiming at the low coding rate problem of the traditional MP algorithm, a coding method of MP atoms was proposed, and the coefficient of the atomic and positional parameters was encoded by using MP atomic energy and position distribution characteristics. The experiment results have shown that: 1)the decomposition method of the block compressed sensing image can effectively reduce the computational complexity of sparse decomposition; 2)that the compression coding method, besides maintaining the strengths of the traditional MP image coding, can effectively improve the coding performance and the encoding rate, and reflect the advantage of the sparse decomposition over the traditional decomposition methods.