← all work classical-computer-vision

Document scanner

[![Python 3.10+](https://img.shields.io/badge/python-3.10+-3776AB?logo=python&logoColor=white)](https://www.python.org/)

Result

Sr	Document	Canny + contour	Otsu + contour	Morph gradient	Saturation (HSV)	Hough lines	minAreaRect
1	motion-blurred text · real	2.7 px	1.6 px	2.0 px	0.6 px	1.7 px	47.9 px
2	till receipt · generated	2.5 px	1.0 px	1.4 px	0.6 px	76.3 px	13.8 px
3	sudoku · real	FAILED	70.1 px	25.3 px	1.0 px	231.3 px	28.0 px
4	ruled form · generated	3.1 px	1.3 px	1.7 px	1.2 px	2.2 px	11.5 px

The finding

What it does

`mermaid flowchart LR A[Photo of a page
angled, cluttered desk, uneven light] --> B[1 · Detect the page
6 methods compared] B --> C[2 · Recover aspect ratio
closed form from perspective] C --> D[3 · Rectify
homography warp] D --> E[4 · Binarise
4 methods compared] E --> F[Flat, readable scan] `

Each stage is scored against exact ground truth, not judged by eye:

---

Elsewhere

Code and full results on GitHub (code private)

Every number is read from the file the run produced. github.com/hammasbuilds · kaggle.com/muhammadhammas13