Image Colorization System

Abstract

In this paper a new method is introduced that leads to colorize a grayscale images. Color plays a vitally important role in the world in which we live. We want to bring new life to old photos by colorizing them. This type of problem normally requires manual adjustment to achieve artifact-free quality, so inspired by the vernal success in deep learning we introduced a technique to automatically colorize grayscale images that combines both global and local image features. Based on Convolution Neural Network, that has been proven able to learn complex mappings from large amounts of training data. In deep neural network has a adjustment layer that allows us to magnificently merge and the local information dependent on small image patches with global priors using the whole image, our architecture can process images of any resolution, unlike most existing outlook based on CNN. We leverage an existing large-scale scene database to train our model to learn the global priors and classify the objects of image to be able to map color to it. We prove our method spaciously on various types of images, combining with black-and-white photography, approx hundred years ago, and show realistic colorizations.