Removal of spatial inconsistencies in automated image colorization using parameter-free clustering and convolutional neural networks

Given a grayscale photograph, this paper tackles the issue of fantasizing about a conceivable shading rendition of the photograph. This issue is underconstrained, so past methodologies have either depended on significant human cooperation or came about in desaturated colorizations. We propose a wholly automated approach that produces lively furthermore, sensible colorizations. This approach utilizes the combination of CNN-based colorization and parameter-free k-means clustering to identify the color spills, so that the image can be recolored to produce color images that are aesthetically more pleasing and plausible than the images produced by the state-of-the-art methods. The performance of the proposed model is evaluated in terms of mean square error and structural similarity and found to be superior to the related works.