A Method for Coloring Low-resolution Black and White Old Movies through Object Understanding

Coloring black and white old movies through video coloring technology is one of the research hotspots in the field of computer vision. Early old movies generally have problems such as low resolution and obscure objects, resulting in unsatisfactory coloring effects. This paper proposes a method for coloring low-resolution black and white old movies through object understanding. The approach first employs Microsoft's COCO dataset to train Conditional GAN (CGAN). The trained CGAN is not only able to learn color, but also able to understand objects. First, the black and white old movie is framed, and then the framed image is sent to the CGAN for coloring, finally color frames generated by the network are synthesized into a movie. The whole experimental process includes object recognition, object understanding and auto coloring. This paper takes the "Waterloo Bridge" movie as an example for testing. The test results show the coloring effects are satisfactory. Therefore, this paper implements the automatic colorization of black and white old movies by recognizing and understanding color objects of low resolution. This research work not only has some theoretical value, but also has the application value.