Colorizing the Past: Deep Learning for the Automatic Colorization of Historical Aerial Images

Abstract

The colorization of grayscale images can, nowadays, take advantage of recent progress and the automation of deep-learning techniques. From the media industry to medical or geospatial applications, image colorization is an attractive and investigated image processing practice, and it is also helpful for revitalizing historical photographs. After exploring some of the existing fully automatic learning methods, the article presents a new neural network architecture, Hyper-U-NET, which combines a U-NET-like architecture and HyperConnections to handle the colorization of historical black and white aerial images. The training dataset (about 10,000 colored aerial image patches) and the realized neural network are available on our GitHub page to boost further research investigations in this field.

Colorizing Multi-Modal Medical Data: An Autoencoder-based Approach for Enhanced Anatomical Information in X-ray Images

“…To address the colorization of historic black-and white aerial photographs, Farella E. M. (et al) introduce a novel neural network architecture called Hyper-U-NET, which combines a U-NET-like design and HyperConnections. Although there were occasional failures when working with low-quality images, the suggested approach generally produced excellent colorization results in many cases [26]. Dias, M. (et al) assessed the effectiveness of a modified W-Net design.…”

“…This method yields good results for many grayscale medical images while retaining the original image's intensity. A healthcare professional may be aided by the colour information as a tool to highlight significant structures by colouring the region of interest [2]. Mathur, A. N. (et al) offers a framework with the intention that it might be applied to multi-modal data in 3D for colorisation while keeping in mind the photorealism of the resulting outputs.…”

“…During colourisation, a vector in the specified colour space used to represent a particular colour is added in place of the old scalar value that represents the pixel's intensity [1]. Some of the most popular colourisation techniques are based on: luminance keying and colour transfer, image analogies, motion estimation, segmentation, colour prediction, probabilistic relaxation, and chrominance blending [2].…”

Relationships between colorization and pseudo-colorization of monochrome images

“…In the case of recolorization, the goal is to replicate the original colors of the image, even though they are unknown to the algorithm, e.g. [3,6,16,18,22,23], etc. Alternatively, the objective is to create realistic-looking hypothetical color versions of grayscale images, representing worlds that might not have originally been in color, e.g.…”

“…learning approaches have become dominant, with neural networks designed to learn color patterns suitable for specific domains, semantics, or content, [3,23]. Some techniques additionally incorporate recognition or learning of image domains or objects to further enhance colorization, e.g., [6,18].…”