H&E Colour Normalisation for Histology Images

A command-line tool to harmonise the colour, tone and saturation of haematoxylin and eosin (H&E)-stained histology images acquired on different microscopes or on different days, making them visually comparable without any loss of spatial resolution.

Method

Colour transfer is performed using the Reinhard et al. (2001) algorithm in the CIE Lab colour space. For each image, the mean and standard deviation of each Lab channel (L, a, b) are matched to a common target computed as the batch average. A tissue mask (based on luminance) excludes background pixels — white areas such as vessels, sinusoids or empty space — from the statistics, so they do not bias the colour reference. No spatial resampling is applied; output images are pixel-for-pixel identical in resolution to their inputs.

Reinhard E. et al. (2001). Color transfer between images. IEEE Computer Graphics and Applications, 21(5), 34–41. https://doi.org/10.1109/38.946629

Requirements

• Python ≥ 3.10
• scikit-image
• tifffile

It is recommended to install dependencies in an isolated conda environment:

conda create -n he-norm python=3.12 -y
conda activate he-norm
pip install scikit-image tifffile

Usage

python3 normalize_he.py INPUT_FOLDER [OUTPUT_FOLDER]

• INPUT_FOLDER — folder containing the .tif/.tiff images to normalise. All files with those extensions found directly inside the folder are processed as a batch.
• OUTPUT_FOLDER — optional. If omitted, a normalized/ subfolder is created inside INPUT_FOLDER.

Example

conda activate he-norm
python3 normalize_he.py ~/histology/june_batch ~/histology/june_batch_normalised

Output files are saved as uncompressed RGB TIFF and named <original_stem>_normalized.tif.

Notes

• The colour target is computed as the mean across the current batch, so results are internally consistent within a run. If you need to match a specific reference image (e.g. a gold-standard slide), the target statistics can be hard-coded in the script by replacing target_mean and target_std.
• The script handles greyscale images (converted to RGB) and RGBA images (alpha channel discarded).
• A UserWarning about a small number of clipped negative Z values during Lab→sRGB conversion is expected and harmless; those pixels are clipped to the valid sRGB range.

Licence

Copyright (C) 2026 Alfonso Mora Corral. GNU General Public License v3.0 — see LICENSE for details.