Abstract: A novel method of image colorization is proposed. It overcomes the problem existed in algorithm of image colorization based on dictionary learning and sparse representation. First,multiple blocks from reference color images was selected according to the sub-contents of target gray-scale image.Second,the multiple blocks are trained respectively to obtain classified dictionaries based on contents.Finally,the best matching sub-dictionary can be decided by minimizing the reconstruction error and the gray-scale image is colorized. This algorithm was automatic and the performance of colorization was improved on the premise of guaranteeing the automation of the image colorization process. Experimental results show that the algorithm is able to colorize the target gray-scale images of different contents automatically.