Automatic Natural Image Colorization
- 300 Downloads
We introduce a technique to automatically colorize natural grayscale images that combines both local and global features. Automatic colorization is a hard problem of computer vision and usually requires user interactions such as human-labelled color scribbles or reference images to achieve proper results. Based on Convolutional Neural Networks (CNN), our model is trained in an end-to-end fashion and can process images of any resolution. We improve the model of Iizuka et al.  by adding edge detection network and adjusting the input of the loss function. We also compare our model with the state of the art and show some improvements. Furthermore, we try colorizing ink wash paintings and achieve a special style.
KeywordsColorization Convolutional Neural Network Self-supervised learning
This research was supported, in part, by Ngoc Dung Beauty Center. We thank members of Ngoc Dung AI Lab for helpful discussions and Duy-Phu Nguyen for his helpful advice.
- 2.Levin, A., Lischinski, D., Weiss, Y.: Colorization using optimization. ACM Trans. Graph. 23(3) (2004)Google Scholar
- 3.Huang, Y.-C., Tung, Y.-S., Chen, J.-C., Wang, S.-W., Wu, J.-L.: An adaptive edge detection based colorization algorithm. In: ACM International Conference on Multimedia (2005)Google Scholar
- 4.Luan, Q., Wen, F., Cohen-Or, D., Liang, L., Xu, Y.-Q., Shum, H.-Y.: Natural image colorization. In: Eurographics Conference on Rendering Techniques (2007)Google Scholar
- 5.Welsh, T., Ashikhmin, M., Mueller, K.: Transferring color to greyscale images. ACM Trans. Graph. 21(3) (2002)Google Scholar