Multi dimensional dataset improves quality assessment of colorized images
MCIQA-2K: A Multi-Dimensional Dataset and No-Reference Quality Assessment Benchmark for Colorized Images
Computer Vision and Pattern Recognition
Summary
Images that are colorized from black and white can look different depending on how they are colored, and there isn’t just one right way. Usual ways to judge the quality of these colorized images don’t match how people actually see them. To solve this, the authors created a large new dataset of 2,000 colorized images and had people rate them based on specific color issues and overall realism. They also built a new tool, called MCIQA, to judge colorized images without needing the original for comparison, and it works better than existing methods on their dataset and others.
What this means in practice
- •For digital imaging engineers: Evaluate and improve colorization algorithms by using a dedicated dataset and tool to better match human perception of color quality.
- •For photo editing software developers: Integrate a no-reference quality assessment tool for automatic feedback on colorized image quality during image enhancement workflows.$Commercial implications: Offers a product feature enabling software to automatically assess and suggest improvements for colorized images, enhancing user experience.
Authors
Yunkai Zhuang, Qihang Yan, Zicheng Zhang, Guangtao Zhai
Abstract
Image colorization is an inherently ill-posed task, since a single grayscale image may correspond to multiple plausible colorized results. Consequently, conventional full-reference image quality assessment (IQA) metrics fail to accurately reflect human perceptual preferences for colorized images. In this paper, we present MCIQA-2K, a large-scale multi-dimensional benchmark specifically designed for no-reference quality assessment of colorized images. We construct a dataset containing 2,000 colorized images generated by five representative colorization models, together with human annotations across three perceptual dimensions: color smearing, semantic color misalignment, and global naturalness. Building upon the proposed benchmark, we further introduce MCIQA, a dedicated multi-branch NR-IQA framework for colorized images. Extensive experiments demonstrate that MCIQA significantly outperforms existing full-reference and no-reference IQA methods on the proposed benchmark, while also exhibiting competitive generalization capability on several widely-used IQA datasets. The dataset and code are publicly available at https://github.com/ARBEZ-ZEBRA/MCIQA.