Metrics for color and contrast differences between images, sorted by whether they need a pixel-aligned reference, with links to code and toolbox implementations. Each entry records how it handles misalignment, whether it sees color, what it outputs, and where to get code.
From left to right, metrics care less about where pixels are. Click a column to filter by alignment, or a topic to filter by both. Numbers are entries per topic in that column.
Tolerating misalignment is not the same as seeing color: check the color label, and run the cast-vs-shift test in the guide before trusting a metric on misaligned pairs.
Six lines of work, with representative entries.
Entries with official code, a toolbox implementation or a third-party implementation. Package names are copied from each repository README (checked 2026-10-11).
| Metric | Year · venue | Topic | Alignment · color | Official code | Other implementations | Status |
|---|
Including work without public code. Within each topic: entries with code first, then standards, then unreleased; newest first within each group.
Download links from each dataset's website or the paper repository.
| Dataset | Year | Source | Content | Size | Used for | Link |
|---|
Color science libraries, IQA toolboxes, standalone difference tools and chart analysis.
A GitHub survey on 2026-10-11 found no curated list dedicated to color metrics; the closest, Awesome-Image-Quality-Assessment, has a two-entry "Color IQA" subsection. Entries here come from the reference lists and citing papers of SPCD, CD-Flow, CD-iNet, MS-SWD, CD-Reasoning, ColorVideoVDP, HyAB, iCID and DeepSSIM, Crossref keyword searches, the evaluation sections of learned-ISP, burst and real-world super-resolution papers, ISO / IEEE camera standards, and common toolboxes.
Paper links were resolved through Crossref, arXiv or the publisher, and repository READMEs were read to confirm official code and package names, on 2026-10-11. Alignment labels follow what each paper claims or what its mechanism implies; “tolerates small shifts” means sub-pixel to a few pixels. A “partial” color label means the paper does not test color.
Known gaps: not a systematic review; IEEE Xplore and ACM DL were not searched exhaustively. Out of scope: no-reference quality models without a color component, color constancy algorithms, colorization and color transfer, and video-only temporal metrics. Found a gap or a wrong label? Contributions are welcome.