Among the conventional algorithms of colorization, color transfer algorithm is the most common one. However, the traversal matching of each pixel is inevitable in the case of target color image. Though existing some improved methods of color correlation based, they make little sense on large resolution images and grayscale video rendering. This paper focuses on the use of CUDA, GPU general parallel computing platform, which is rapidly developed in recent years, accelerating colorization and adopting Secondary Coloring method to eliminate the effects of noise. The experimental results imply better acceleration.