To solve the problems of fuzzy phenomenon and pseudo color in the denoising process of color images, a dictionary algorithm combining multiple information is proposed. Firstly, the definitions of the weighted gradients based on the channel model values in RGB color space are presented. On this basis, an over-complete structure of the dictionary is established, which uses brightness, weighted gradient and color information. Secondly, the iterative dictionary training process is continually updated and processed by using the sparsity of the noised image, and the optimal sparse coefficients and optimal learning dictionary are found, which can separate noise information from useful information of images, accurately reconstruct images which only requiring the color values computation and obtain the denoised color image. Experimental results show that, compared with the existing algorithms, the proposed algorithm achieves better visual effects and higher objective index values under different noise intensities, indicating that the algorithm has good denoising performance.