Work
Computer Vision & OCR
Urdu Document OCR Preprocessing
Computer Vision Pipeline for Degraded Complex Script Recognition
- By
- Muhammad Haider Akhtar
- Field
- Document Intelligence / AI
- Time
- Computer Vision Engineering
- Kind
- Founder's Professional Experience
The problem
Historical and degraded documents in cursive Nastaliq and Urdu scripts suffer from extremely high OCR character error rates due to paper yellowing, bleed-through ink, uneven illumination, and rotational skew.
The work
Developed an automated computer vision preprocessing engine utilizing Sauvola adaptive thresholding, morphological noise filtering, and Hough transform skew correction to dramatically increase downstream OCR character accuracy.
- Python
- OpenCV
- Tesseract OCR
- NumPy
- Adaptive Thresholding
- Skew Correction
What it does
- Substantially boosted recognized character accuracy on heavily degraded scanned archives without requiring neural retraining
- Engineered high-throughput batch processing handling multi-page PDF documents in parallel
- Created reproducible benchmark comparisons between standard Otsu binarization and specialized adaptive methods
Technical research and engineering project conducted by founder Muhammad Haider Akhtar.
NextAurygen Technologies PortalWeb Application