PALETTE BASED IMAGE RE-COLORIZATION USING NEURAL NETWORKS

Abstract

Traditionally or Manually grey scale image colorization requires a lot of time and engage user to consider very much minor details in the form of positioning various colour scribbles, performing image segmentation or visualizing at similar pictures even effective software like photoshop can take about month to colorize a black and white image. This is all because of the vast range of shades present in the picture. Colour of the pixel are highly dependent on the features of its neighbours convolutional neural networks. By working on a dataset of different objects like sky, man, cats, dogs, bulls etc.Solving the problem by converting them into black and white 256x256x1 arrays which contain only the light values of the images in the Lab colour schema our model output 256x256x3 arrays contain blue-yellow, green-red and white values .after going through several CNN having different convolutional and training size we tried to build a more advance model that is comprise of the Vggl6 architecture and autoencoders which give us more ppossible results.