Reproduce “Palette-based image decomposition, harmonization, and color transfer” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jianchao Tan, Jose Echevarria, Yotam Gingold
A new palette-based framework unifies image decomposition, color harmonization, and color transfer, employing a highly efficient geometric algorithm in 5D RGBXY-space that decomposes 6 Megapixel images in 20 milliseconds. Extensive perceptual studies reveal that while harmonized palettes are often preferred, original images are generally favored over algorithmically harmonized ones, challenging some traditional color theory assumptions.
Key insights
- Embedding color and spatial information into a 5D RGBXY-space enables an extremely efficient and spatially coherent geometric image decomposition method, eliminating the need for slow numerical optimization.
- Perceptually uniform color spaces, such as CIE LCh, are critical for developing robust and intuitive color harmonization models that align with human visual perception.
- Large-scale perceptual studies reveal that while algorithmically harmonized color palettes are often preferred, applying global harmonization to complex images can lead to a preference for the original, suggesting the importance of context and user expectation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.