A Python analysis pipeline that measures true 24-bit RGB Shannon entropy, compares entropy ordering with compressed file size, and builds an interactive static results site.
The project turns an information-theory formula into a reproducible data pipeline: recursive ingestion, exact color-frequency analysis, deterministic ranking, aggregate export, and an explorable report. It also illustrates that image complexity and compressed file size are related but not interchangeable measurements.
- Recursively load JPEG images.
- Count exact RGB color occurrences.
- Calculate Shannon entropy:
H(X) = -sum(p(x) * log2(p(x))). - Sort images by entropy and compare the ordering with file size.
- Export aggregate CSV results and build a browsable HTML report.
python -m venv .venv
pip install -r requirements.txt
python src/compute_entropy.py --root data/images --out_csv results/entropy.csv
python src/build_site.py --csv results/entropy.csv --site_dir reports/site
The ranking view is generated from the committed lowest10.csv and highest10.csv outputs and does not redistribute third-party images.
In the analyzed collection, observed entropy ranged from approximately 2.79 bits for a visually simple image to 17.38 bits for a complex autumn-foliage image. Aggregate results are included, while the third-party source images are not redistributed.
The analysis code and aggregate outputs are published; the third-party source-image collection and generated thumbnails are excluded.
CSCI 49000 AIT — Artificial Intelligence for IoT, Fall 2025.
Built by Ahmed Balde as a Python data-analysis and information-theory project. See more work on GitHub.