Automatic Colorization of images using Auto-encoders

Abstract

Colorization of images is one of the preliminary steps of image analysis and documentation. Autocolorization is an automated process of converting a single-channeled image into a complete colorized 3 channel RGB image. There has been extensive research gap in the field since the dawn of deep learning. This document is a model for a statistical-learning driven approach to approach Autocolorization through building an Encoder-decoder model with Convolutional neural networks. Learning Models: Keras, openCV, Numpy and Tensorflow. A direct function to convert grayscale into coloured images that can be coupled with various software or sensors. The results obtained provide a visualization of autocolorization with different regularization techniques and optimizers.