Extraction of Approval Stamps from Engineering Drawings
The Challenge
Engineering doc controllers add IFR codes and revision comments freely anywhere in technical drawing documents. Regular expressions fail without fixed anchors — making automated extraction of approval stamps and engineering annotations impossible with rule-based approaches.
What We Built
A Computer Vision + OCR pipeline performs intelligent document manipulation regardless of annotation position. HSV pixel mask generation separates human annotations from the base drawing. OCR applied to segmented layers produces structured JSON output capturing every stamp and technical comment.
Operational Impact
Automated extraction of approval workflows previously blocked by visual document complexity. Full auditable digital record of engineering decisions and drastic reduction in delivery cycle bottlenecks.
Technology Used
Python
PyTorch
OpenCV
Azure Custom Vision
FastAPI
