To pursue the spatial consistency in the process of colorization, unpleasing transferred color may be produced due to over-smoothing. To solve this problem, the authors propose a colorization approach based on an adaptive weighted average method to assign color from a reference color image to a target grayscale image. In this approach, support vector machine and improved simple linear iterative clustering are combined to obtain class probability distribution and classification labels with high local spatial consistency. According to the classification results, the color candidate in reference image is sought by space features matching with the corresponding class labels. Thereafter, an adaptive weighted average filter based on the class probability is applied to compute the chrominance value by averaging the neighborhood of the candidate pixel, and then the color is assigned to the corresponding pixel in grayscale image. The adaptive weight filter can ensure spatial coherency and avoid over-smoothing simultaneously. Experimental results demonstrate the superiority of our method compared with the existing methods.