Deep-Learning Based Colorization of Grayscale Images and Videos

The area of automatic colorization of grayscale images and videos is an important area of research that has attracted a lot of attention in recent times because of the advancements in deep learning techniques. The paper proposes a unified framework for colorizing images and videos based on the implementation of a Conditional Generative Adversarial Network and a U-Net-based generator. The colorization of images and videos is performed in the Lab* color space, where the luminance channel is used to colorize the output. The proposed method differs from other existing colorization techniques in that it uses a single model to process both images and videos, whereas other techniques require different models to process these two different types of inputs. The colorization of the video is performed by processing each frame of the video separately and then reconstructing the output video by arranging the processed frames in sequence. The proposed method does not require any reference images and can be performed efficiently.