ClearSheet

ClearSheet is a full-stack document digitization product built to take a raw phone photo of a page, detect the document automatically, flatten it, enhance it into a scan-style image, and extract text with OCR. The project combines a modular OpenCV pipeline, a FastAPI backend, and a polished Next.js frontend for upload, review, and export.

The backend is intentionally split into small stages so each part of the workflow is testable and easy to evolve. Preprocessing prepares edges and a document mask, contour detection searches for a stable four-point page boundary, the transformer warps and auto-rotates the sheet, and post-processing builds the final scan before OCR ranking chooses the strongest text result.

• ✓Preprocessing normalizes contrast, applies blur/Canny, and builds a cleanup mask for document candidates.
• ✓Document detection prefers a convex quadrilateral, then falls back to minAreaRect when the page edges are imperfect.
• ✓Perspective correction orders corners, warps the page, and estimates residual tilt with Hough-line based auto-rotation.
• ✓Three scan profiles (clean, balanced, ocr-optimized) tune CLAHE, sharpening, thresholding, and text expansion differently.
• ✓OCR is evaluated across multiple image variants and Tesseract configs, then ranked by confidence and useful content length.

The frontend wraps the backend with a practical workspace: upload a source image, choose OCR languages and scan mode, inspect optional debug snapshots, review the transcript, and export the result without leaving the page.

• •Upload a raw photo from desktop or mobile and send it to the FastAPI scan endpoint.
• •Choose fra, eng, or fra+eng plus the scan profile that best fits readability or OCR.
• •Inspect optional debug images for edges, contour detection, and the warped document.
• •Review extracted text, OCR confidence, and tokenized text geometry returned by the API.
• •Download the cleaned scan as PNG or generate a searchable PDF directly in the browser.

• •The backend is organized into dedicated preprocessing, detection, transform, post-processing, and settings modules instead of one monolithic script.
• •Settings auto-detect local Tesseract and bundled tessdata paths, which makes Windows development smoother and Docker deployment cleaner.
• •docker compose up --build starts the frontend and backend together for a reproducible full-stack local environment.
• •Backend tests cover health and scan endpoints, invalid uploads, document-not-found cases, transformer behavior, post-processing, debug images, and optional OCR integration.

• Live app ↗Next.js frontend deployed on Netlify
• GitHub repository ↗Full ClearSheet source code
• Backend API ↗FastAPI service deployed on Hugging Face Spaces