In order to improve the reconstruction quality of compressed sensing images ， a band-CS (Compressed Sensing) algorithm with smooth projection Landweber is proposed on the basis of discrete cosine transform (DCT). Since different DCT coefficient bands have different effects on the final image reconstruction quality ，first ， this algorithm divides an image into several blocks and carries out DCT for each block. Secondly ， DCT coefficients are regrouped according to the band energy ， and the band with higher energy is sampled by random matrix with greater sampling rate. Then ， the blocking effect is eliminated by using a smoothing filter. Finally ，the compressed sensing reconstruction of images is implemented by means of smooth projection Landweber.Experimental results show that ， in comparison with BCS-SPL and MS-BCS-SPL algorithms ， the proposed algorithm improves the peak signal-to-noise ratio of reconstructed images significantly.