A compressed sensing reconstruction method based on nonlocal similarity, low rank matrix and minimum total variation (TV) is proposed, considering the non-local similarity of remote sensing images. It fully exploits the nonlocal similarity prior, local smoothness prior of remote sensing images and the low rank properties of matrix. A new joint block matching method based on Euclidean distance and structural similarity is developed, which makes the matching result more accurate. The reconstruction of high quality remote sensing image is realized finally. Simulation results confirm that the proposed algorithm can achieve high reconstruction quality comparing with the traditional reconstruction method based on sparse transform domain or TV regularization. The peak signal to noise ratio and structural similarity have a great improvement，and the effectiveness of the proposed method is verified.