Metrics for commercial OCR solutions

[bora-skylight]

Acceptance Criteria

• Set up AWS Tesseract & Azure Vision AI

• measure E2E scanned metrics for two identical forms (one digitally filled, one hand-filled)

How long does an e2e form run take Measure the accuracy/precision/recall OCR Model: Measure the accuracy/precision/recall/time of the following

Scanned typed/handwritten text Scanned typed/handwritten numbers Scanned typed/handwritten phone numbers different formats 555-555-5555 (555) 456-4567 Scanned typed/handwritten addresses

Steps

1. Create AWS Tesseract and Azure AI vision instance(API key) and run locally
2. Ensure formatting of ground truth matches formatting of ocr result
3. Add workflow that compares the results based on the above metrics