Compressed sensing system can reconstruct the original image from fewer measurements using the sparse priors of image. Current research in compressed sensing has devised algorithms for grayscale images, but there are few methods for color images. Since each of the color channels is highly correlated, the result of simply extending the reconstruction algorithm of grayscale images to three channels of color images is not satisfying. Aiming at improving the reconstruction quality of color images, compressed sensing of color images based on local Gaussian model in the dual-tree complex wavelet is proposed, which uses the dual-tree complex wavelet having the property of translation invariance as the sparse representation of natural images. Priors of the inter-cross correlation of three channels of color images and the local neighbor statistic distribution of the wavelet coefficients are applied in reconstruction. Experimental results show that the proposed algorithm can improve the peak signal-to-noise ratio and the visual quality.