Document Block Segmentation for OCR

[lancelafontaine]

Description As a user, I would like the system to extract the handwriting for a picture of a lease.

This task should represent the remainder of the work required to implement OCR of leases and extraction of facts from lease values.

Issue #263 implemented a proof-of-concept for handwritten character recognition, which still requires integration into the task_service API

Issue #25 integrated the alignment of a lease document from an arbitrary picture of a lease into the task_service API.

Scope of Work

• [ ] Use Tesseract to extract images from segments of a lease
• [ ] Extract words from these images using the work done in #263
• [ ] Integrate code changes into the task_service API

Demo Requirement and Acceptance Criteria

1. The user is able to upload a photo of a lease
2. A new aligned document image is retrieved based on the original photo (done in #25)
3. Relevant segments of the lease (checkboxes, handwritten textboxes) are extracted in image format
4. Images are fed into character recognition CNN (PoC complete but requires integration in #263)