ImageProcessing — Crop Utilities

Purpose

This repository contains utilities for extracting regions of interest from scanned documents or images (age/gender fields and tight crops around foreground content). It helps preprocess images for downstream OCR or classification models.

Features

• Deterministic crops for age and gender regions.
• Tight crop around dark foreground pixels (suitable for scanned digits/text).
• Robust validation and safe image slicing behavior.

Tech Stack

• Python 3.10+
• OpenCV (cv2)
• NumPy

Setup

1. Create and activate a virtual environment:

python -m venv .venv
source .venv/bin/activate

2. Install dependencies:

pip install -r requirments.txt

3. Run quick interactive test (example):

from blocks import crop
import cv2

img = cv2.imread('data/pdfs/example.png')
age_region = crop.crop_age(img)
gender_region = crop.crop_gender(img)
cropped = crop.cut2cut_crop(img)

Configuration

There are no runtime secrets required for the cropping utilities. If you extend the project (for model inference, cloud storage, etc.) list env vars and secret management here.

Folder Structure

• blocks/ : image-processing modules (including crop.py).
• data/ : sample inputs and outputs.
• digits/ : templates and digit-related resources.
• notebook_archive/ : exploratory notebooks.

Known Limitations

• The deterministic crop ratios (crop_age, crop_gender) are tuned for a specific document layout and may fail for other templates.
• cut2cut_crop assumes a light background and darker foreground; it will not work well with noisy or multi-colored backgrounds.

Next Steps to Make This Project More Contributable

• Add CONTRIBUTING.md with contribution guidelines and CODE_OF_CONDUCT.
• Add unit tests for blocks/crop.py and CI (GitHub Actions) to run them.
• Add a LICENSE file (e.g., MIT) to enable open-source contributions.
• Add type-checking via mypy and a linter config (e.g., flake8/ruff).

---

See src/crop.py for detailed documentation of the functions.