Image Colorization: A Survey and Dataset

TL;DR

This survey addresses the rapid development of deep learning-based image colorization by presenting a taxonomy of seven classes, introducing a colorization-specific Natural-Color Dataset (NCD), and benchmarking state-of-the-art methods. It covers plain networks, user-guided methods, domain-specific approaches, text-conditioned colorization, diverse outputs, multi-path architectures, and exemplar-based techniques, highlighting architectural choices, losses, and training protocols. The paper reveals trends such as GANs enabling diverse colorizations and emphasizes the need for better evaluation metrics, standardized datasets, and open-source benchmarks to drive progress. It also discusses limitations in generalization to complex scenes, and proposes future directions including unsupervised learning and attention mechanisms to improve robustness and realism in colorization outputs.

Abstract

Image colorization estimates RGB colors for grayscale images or video frames to improve their aesthetic and perceptual quality. Over the last decade, deep learning techniques for image colorization have significantly progressed, necessitating a systematic survey and benchmarking of these techniques. This article presents a comprehensive survey of recent state-of-the-art deep learning-based image colorization techniques, describing their fundamental block architectures, inputs, optimizers, loss functions, training protocols, training data, etc. It categorizes the existing colorization techniques into seven classes and discusses important factors governing their performance, such as benchmark datasets and evaluation metrics. We highlight the limitations of existing datasets and introduce a new dataset specific to colorization. We perform an extensive experimental evaluation of existing image colorization methods using both existing datasets and our proposed one. Finally, we discuss the limitations of existing methods and recommend possible solutions and future research directions for this rapidly evolving topic of deep image colorization. The dataset and codes for evaluation are publicly available at https://github.com/saeed-anwar/ColorSurvey.

Figures (10)

• Figure 1: Taxonomy: Classification of the colorization networks based on structure, input, domain, and type of network.
• Figure 2: Plain networks are the earlier models with convolutional layer stacking with no skip or naive skip connections.
• Figure 3: User-Guided networks are the ones that require a user to input the color at some stage of the network during colorization.
• Figure 4: Domain-Specific Colorization networks colorize images from different modalities such as infra-red, radar images.
• Figure 5: Text-based colorization networks are based on the text input with the grayscale image.