A Novel Approach to Image Colorization Algorithm using Convolutional Neural Networks

• Published in: 2022 International Conference on Applied Artificial Intelligence and Computing (ICAAIC) pp. 243 - 250
• Main Authors: Singla, Manav, Jain, Aanvi, Mantri, Manav, Tulshain, Shobhit, Muthu, Rajesh Kumar, Naidu, Rani Chinnappa
• Format: Conference Proceeding
• Language: English
• Published: IEEE 09.05.2022
• Subjects: CNNs; Gray-scale; Image color analysis; Image Processing; Image segmentation; Linear programming; Neural Networks; Optimization; Prediction algorithms; Python; Representation learning; Training
• DOI: 10.1109/ICAAIC53929.2022.9793234

This work tackles the topic of imagining a noticeable color rendition of a picture using a grayscale image as input. Because this issue is plainly unconnected, earlier solutions have generally depended on extensive user engagement or resulting in underexposed clarifications. In this paper, a fully automated method is developed to create rich as well as lifelike colors. It presents the challenge as a classification job to recognize the intrinsic ambiguity of the problem and employ rearranging classes throughout the training period to improve the variety of colors within results. At the time of testing, the system was set up as a feed-forward pass to Convolutional Neural Network (CNN) as well as trained on over one million color pictures. It marks the program by utilizing the "algorithm Turing test," which asks individuals to tell the difference between our program's graphics and real-world photos. Our approach effectively deceived people in 32% of tests, which is far higher than prior methods. The CNN will be used to enhance grayscale conversion as well as auto colorization algorithms. Predicting brilliant and realistic colors will be done using a feed-forward, two-phase system based on CNN.