A Novel approach for Gray Scale Image Colorization using Convolutional Neural Networks

Image Colorization is the process of coloring a grayscale image is done by using a black and white known as grayscale image as input and obtaining the output in RGB format simply known as color image is called Image Colorization. The technique of colorizing an image has more impact on different fields such as photography of astronomical objects, visuals of electronic microscope and CCTV surveillance. To build an automated system for color gray scale image use Deep learning algorithm. In this paper the analysis of convolutional neural networks (CNN) is performed. In convolution neural networks, during the training phase of the model the features are extracted by the feature extractor - Inception ResNet V2, a pre-trained model developed and released by Google (trained on huge dataset of nearly 1.3Million images). Now the features extracted by the feature extractor are combined with the output of the encoder at fusion layer. Fusion layer is responsible to train the model with features and decoder helps to do up-sample and gets the required output. The epoch and steps per epoch are compared based on performance measures from the error rate. The epoch with the least error rate is considered as the best one to be used to colorize grayscale image. From this work the best epoch and steps per epoch can be identified. Thus the use of convolutional neural networks to colorize grayscale images shows better performance.