The colors in a watercolor painting are transparent. The watercolor style uses one layer of color over another to achieve a special transparent visual effect. Therefore, the watercolor pixel color change level is more complex, and the color edge is not apparent. The traditional image processing method is challenging to carry out high-quality image transformation, image coding, image compression, and image segmentation. This paper proposes a watercolor image processing method oriented to big data analysis. First, the collected watercolor image is preprocessed to eliminate background noise in the watercolor image. Then, the watercolor image is colorized through image semantic segmentation. Based on the semantic segmentation results, the images can be clustered by different layers. These layers are then colorized according to their properties. Finally, the colored layers are combined with a degree of transparency to get the final watercolor-style image. The watercolor obtained by this method is more consistent with the actual watercolor painting form and fully uses the advantages of big data analysis. The computer simulation of the watercolor painting process and the parallel deep learning method can significantly improve the algorithm's efficiency.