Exposure Fusion

Python implementation of exposure fusion of multiple images, using the algorithm described in:

Mertens, Tom, Jan Kautz, and Frank Van Reeth. "Exposure Fusion." Computer Graphics and Applications, 2007. PG'07. 15th Pacific Conference on. IEEE, 2007.

Combines a bracket of differently-exposed images into a single well-exposed result via multi-resolution pyramid blending, without HDR radiance map estimation or tone mapping.

Features

• Exposure fusion — Laplacian pyramid blending with contrast, saturation, and well-exposedness weighting
• Automatic alignment — Translation-only alignment via ECC (Enhanced Correlation Coefficient), pure NumPy
• Time-decay weighting — Progressive weighting for sequential image stacks (e.g. time-lapses)
• No OpenCV required — All image operations reimplemented in NumPy; runtime deps are numpy + Pillow only

Requirements

• Python >= 3.10
• numpy
• Pillow

OpenCV is optional and only needed for running the test suite.

Installation

pip install exposure_fusion

Or from a local checkout:

pip install .

With test dependencies:

pip install "exposure_fusion[test]"

CLI Usage

exposure-fusion [-h] [-o OUTPUT] [-d DEPTH] [--time-decay TIME_DECAY]
                [--align] [-v] IMAGE [IMAGE ...]

Arguments:

Examples:

exposure-fusion samples/peyrou_mean.jpg samples/peyrou_under.jpg samples/peyrou_over.jpg -o result.jpg
exposure-fusion --align -d 4 samples/peyrou_mean.jpg samples/peyrou_under.jpg samples/peyrou_over.jpg -o result.jpg
exposure-fusion --time-decay 4 samples/time_decay_1.png samples/time_decay_2.png samples/time_decay_3.png samples/time_decay_4.png -o fusion.png

Also invocable as python -m exposure_fusion.

Python API

Pass file paths directly to exposure_fusion and align_images — no manual I/O needed:

from exposure_fusion import exposure_fusion, align_images, load_image, save_image

# Load bracket exposures
img1 = load_image('samples/peyrou_mean.jpg')
img2 = load_image('samples/peyrou_under.jpg')
img3 = load_image('samples/peyrou_over.jpg')

# Optional alignment
images = align_images([img1, img2, img3])

# Fuse
fusion = exposure_fusion(images, depth=4)

save_image('samples/peyrou_fusion.jpg', fusion)

# Or pass paths directly — no load_image needed:
fusion = exposure_fusion([
    'samples/peyrou_mean.jpg',
    'samples/peyrou_under.jpg',
    'samples/peyrou_over.jpg',
], depth=4)
save_image('samples/peyrou_fusion.jpg', fusion)

# Time-decay fusion (e.g. time-lapse)
images = [load_image(f'samples/time_decay_{i}.png') for i in range(1, 5)]
fusion = exposure_fusion(images, depth=3, time_decay=4)
save_image('samples/time_decay_fusion.png', fusion)

Tests

pip install "exposure_fusion[test]"
pytest tests/

License

MIT