The Challenge
The existing OCR pipeline was checking word containment in article boxes using inefficient polygon operations, costing $325K annually in AWS Lambda costs with slow processing times.
Our Solution
Implemented a ray-tracing algorithm from computer vision to quickly determine word-to-article containment relationships, reducing computational complexity from O(n²) to O(n log n).
Technologies Used
Results & Impact
- Reduced annual processing costs from $325K to $11K
- Improved processing speed by two orders of magnitude
- Enabled real-time OCR processing for large document batches