Image Colourization (using Machine Learning)

Computer Science & Engineering

Colorization of grayscale images has become a more studied area in recent years, due to the advent of Deep convolutional neural networks. We are going to apply this concept to colouring black & white images through deep learning. While Previous similar studies have focused primarily on natural image colouring, cartoons colouring has traditionally been done by using hand drawing Techniques. The proposed method is a Self-Supervised Learning or Fully Automated Approach. The main goal of the project report is to provide an overview of how to convert a grayscale image to a colourful image using colorization problems. achieving artifact-free quality usually requires manual matching which is considered as a very difficult problem. This process usually requires careful selection of colourful suggestive images. Given a grayscale photo as input, this article addresses the problem of hallucinating a photorealistic colour version of the photo. This issue is obviously not limited, so previous approaches have either relied on significant user interaction or resulted in desaturated results. We offer fully automatic approach to providing vivid and realistic colorizations. We accept the underlying uncertainty of the problem by posing it as a classification task and use class rebalancing during training time to increase the diversity of colours in our result. The system is implemented as a feed-forward pass on a CNN during test time and is trained on over a million colour images. Evaluate the algorithm with a "Turing test", by asking participants to choose between the generated colour image and the original colour image

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.