SENSOR DISTANCE LEARNING FOR CROSS-CAMERA COLOR CONSTANCY

Authors: Rafique Ahmed, Damien Muselet, Philippe Colantoni, Alain Trémeau

Four-step illuminant estimation and von Kries color correction

Supplementary Material for ICIP25 Paper
1. Select any scene below
2. Wait until image is white balanced
3. Click download if you want to save the white balanced image
1
Input Image
Demo Image

📸 Select an Image:

Loading available images from GitHub...
2
Log-Chroma Histogram

Computing histogram...

3
Color Contrastive Network

Analyzing histogram...

4
von Kries Corrected

Applying correction...

🧠 Our Method

📊 Log-Chroma Analysis

Convert RGB to log-chroma space and compute 128×128 histogram with UV coordinate channels for robust illuminant estimation.

🔥 Color Contrastive Network

Multi-encoder convolutional cross-correlation network trained using our novel color metric and Siamese approach processes histograms to estimate illuminant.

🎯 Illuminant Estimation

Network outputs probability heat map P, from which the peak location gives the estimated illuminant color.

⚖️ von Kries Correction

Apply diagonal chromatic adaptation transform to normalize the image under the estimated illuminant.