This paper addresses the problem of image colorization (adding color to grayscale images), a topic of active research in the field of computer vision. Colorization is a highly uncertain problem, as two of the three image dimensions are lost. However, we emphasize that scene semantics and surface texture provide important clues to color. Instead of traditional regression methods, we explore automatic image colorization using classification and adversarial learning. Building on previous research, we build and refine models, and then compare and analyze them.
