blood-group-fingerprint-simple

Simple blood-group fingerprint classifier (single-file)

One Python file that runs the whole classical-ML pipeline end to end:

load BMPs  ->  preprocess (CLAHE + Gabor)  ->  features (HOG/LBP/GLCM)
            ->  stratified 70/15/15 split  ->  train 6 classifiers
            ->  evaluate on held-out test  ->  save models + results.json

Run

pip install -r requirements.txt
python main.py

First run takes ~3 minutes (preprocesses + extracts features for all 6,000 images, then trains six models). Subsequent runs reuse the feature cache and finish in ~1 minute.

Outputs

• cache/features.npy, cache/labels.npy — the cached feature matrix
• models/<name>.pkl — six trained sklearn pipelines
• results.json — leaderboard with per-class precision/recall/F1

Models trained

Honest caveat

The dataset's filenames are cluster_N_*.BMP where N matches the class — strong evidence the labels were generated by clustering rather than real ABO+Rh blood typing. A linear model on handcrafted features matches a deep network on this data, which means both are fitting a dataset artifact rather than biology. Do not use this for medical decisions.