This project takes an input grayscale image and produces the output as a colourized image. It works based on a deep learning method. LAB colour space is used where ‘L’ lightness intensity is the input and the network is trained to predict ‘a’ and ‘b’ channels, which are green-red and blue-yellow colour balances, respectively.
This project showcases a Convolutional Neural Network (CNN)-based deep learning model designed to bring grayscale images to life by transforming them into vibrant and realistic colorized versions. The project leverages the LAB color space, utilizing the 'L' channel for lightness and predicting the 'a' and 'b' channels for color balance.
Image Colorization using CNN is a testament to the capabilities of deep learning in the domain of computer vision. It brings monochrome images into the world of colors, breathing new life into them. This project goes beyond mere colorization; it represents the culmination of extensive research and training to surpass the 32% benchmark, often regarded as the "colorization Turing test." It is not just for modern photos but extends its magic to restoring legacy black and white images.
-
Deep Learning Excellence: Our CNN model has been meticulously crafted and trained to excel in the task of image colorization, far surpassing traditional benchmarks.
-
LAB Color Space: By using the LAB color space and predicting 'a' and 'b' channels, we ensure that the colorization results are vivid, visually pleasing, and natural.
-
Legacy Image Colorization: Our model is versatile and capable of colorizing contemporary and historical images, preserving their original charm.
Original
Colorized
Original
Colorized
