Add Block Blast OCR - Training-free geometric OCR for structured grids

[dffge552]

Hi @kba,

I've developed a novel OCR approach that doesn't use ML models and would like to contribute it to awesome-ocr.

Project: Block Blast OCR
Repo: https://github.com/dffge552/block-blast
Live Demo: https://block-blast01.netlify.app/

Key Innovation:

• Training-free OCR using perspective transform + bilinear interpolation
• 99.5% accuracy for structured 8×8 grid recognition
• No ML model required - pure geometric approach
• User-guided calibration (5 points) instead of automated feature detection
• <100ms processing time

Technical approach:

1. User marks 4 corners + 1 empty cell reference
2. Homography transformation to normalize perspective
3. Adaptive color thresholding based on reference point
4. Direct cell-by-cell state extraction

This is different from traditional template matching or ML-based OCR. It's optimized for structured data recognition where the layout is known but perspective/lighting varies.

Suggested category:

• OCR Engines (as a non-ML alternative)
• Or a new "Geometric OCR / Non-ML OCR" subsection?

Would this be a good fit for awesome-ocr? Happy to submit a PR if you think it belongs here.

Thanks!