Superpixel based Image Colorization with Automated Reference Image Selection

Abstract

Colorization has received a lot of attention recently. The use of colorization can revitalise dated images, videos, and films. In scientific and medical photographs, colorization can be used to draw attention to certain details. If we can transmit compact grayscale data and restore it to colour on the receiver, we can lower the expenses associated with sending photos and movies over the network. Colorizing images has been tried using scribble-based, example-based, and deep learning techniques over the years. Scribble-based and example-based coloring requires manual supervision for selecting the right color for image regions and for the selection of reference images for color transfer. Deep learning techniques are self-sufficient, but they must be trained on a large corpus of image data to colourize efficiently. We propose a variant of the color-by-example technique in this paper. Normally, the performance of color-by-example relies heavily on the selection of reference images for colouring, necessitating the use of a human operator. We propose a technique for automated reference image selection that employs a hybrid texture matching model to select the best reference image for colouring the given query image. A superpixel-based method is used for transferring the color from the reference image to the query image. The proposed method performs well at identifying appropriate reference images and producing visually appealing coloured images. © 2023 IEEE.