🎨 Palette-based Color Harmonization via Color Naming

Danna Xue^1,2, Javier Vazquez-Corral², Luis Herranz³, Yanning Zhang¹ and Michael S. Brown⁴

¹Northwestern Polytechnical University, ²Computer Vision Center, Universitat Autònoma de Barcelona, ³Universidad Autónoma de Madrid, ⁴York University

This is the official implementation of the paper "Palette-based Color Harmonization via Color Naming" published in IEEE-SPL

Repo for the paper: Palette-based Color Harmonization via Color Naming

Overview

Working from a color-naming model, we derive a set of prototype colors and demonstrate that mapping an image's extracted color palette to the nearest prototype colors effectively harmonizes the image's colors.

All colors in the color space are categorized into 11 classes by a color-naming model, and colors with the same color name are clustered. The color with the highest saturation in each cluster is selected as a color in the prototype palettes. Given an input image, the source palette of the image is extracted. Then, the color name of each color in the source palette is identified by the same color-naming model. By searching for the color with the smallest difference from the source color in the prototype palette with the same color name, the target color is obtained. Finally, the image is recolored based on the target palette to achieve color modification.

Dataset:

• MIT-Adobe 5K dataset: Download the .DNG files

We used the Camera Raw tool in Photoshop, which preprocesses image colors based on the metadata of the raw images, automatically selecting configuration files such as ACR 4.4, ACR 4.6, or Adobe Standard depending on the camera model. This preprocessing method produces images with more vibrant colors, consistent with the sRGB images saved directly from camera shots. We test on the last 500 images.

• PPR10K dataset: Download the source .tif files

We use the validation split of the source 360p 16bit sRGB images and convert them to 8 bits.

Code:

Run experiment:

• How to run the code:

python main.py --data_dir "your_test_image_folder" --save_dir "your_results_folder"

Acknowledgement:

Palette extraction and recoloring parts of the code are inspired from the matlab code multi-image-color-consistency.

The metrics calculation are from: NIQE, BRISQUE, Color Harmonization.

Color histogram visualization are from: Color Histogram.

We thank the authors for providing the source code.

BibTeX

If you find our work useful in your work, please consider citing us.

@article{xue2024palette,
  author={Xue, Danna and Vazquez-Corral, Javier and Herranz, Luis and Zhang, Yanning and Brown, Michael S.},
  journal={IEEE Signal Processing Letters}, 
  title={Palette-Based Color Harmonization via Color Naming}, 
  year={2024},
  volume={31},
  pages={1474-1478},
  doi={10.1109/LSP.2024.3401612}}