Document Image Skew Estimation

example workflow

Cover Image

Table of Contents

Document Image Skew Estimation

Table of contents generated with markdown-toc

Installation

pip

pip install jdeskew

conda

conda create --name jdeskew -c conda-forge jdeskew

How-to-use

using python

from jdeskew.estimator import get_angle
angle = get_angle(image)

from jdeskew.utility import rotate
output_image = rotate(image, angle)

Docker

https://hub.docker.com/r/phamquiluan/jdeskew/tags

# build 
DOCKER_BUILDKIT=1 docker build -t jdeskew .

# run
docker run -p 8000:80 jdeskew

# test
curl -v -F file=@sample.png localhost:8000/predict

using cog

https://github.com/replicate/cog

cog build --debug
cog predict -i input=@skew.png

# Output:
# Running prediction...
# {
#   "angle": -0.12520868113522532
# }

Download Paper

Link1: https://ieeexplore.ieee.org/document/9897910

Link2: https://www.researchgate.net/publication/364320913_ADAPTIVE_RADIAL_PROJECTION_ON_FOURIER_MAGNITUDE_SPECTRUM_FOR_DOCUMENT_IMAGE_SKEW_ESTIMATION

Link3: https://huggingface.co/papers/2603.05942

Performance Comparison on DISE 2021

CE: Correct Estimation rate

WE: Worst Error

	AED	TOP80	CE	WE
FredsDeskew	10.82	0.09	0.54	109
PypiDeskew	16.59	0.24	0.2	141
Koo, Hyung Il et al.	0.22	0.09	0.48	9.43
CMC-MSU	0.27	0.11	0.43	23.2
LRDE-EPITA-a	0.14	0.06	0.66	10.61

Our (1024)	0.11	0.07	0.67	1.13
Our (1500)	0.09	0.05	0.78	1.13
Our (2048)	0.08	0.04	0.84	1.13
Our (3072)	0.07	0.04	0.86	1.13
Our (4096)	0.08	0.04	0.83	1.18

DISE 2021 Dataset

This datasets are built upon three other datasets: DISEC 2013, RVL-CDIP, RDCL 2017. So I urge you to respect their LICENSE.

Dataset Name	URL
DISE 2021 (45 degree)	https://drive.google.com/file/d/1a-a6aOqdsghjeHGLnCLsDs7NoJIus-Pw/view?usp=sharing
DISE 2021 (15 degree)	https://drive.google.com/file/d/1BLiuu-j28dbuPFi4n3C0KuV6vXGmB0qS/view?usp=sharing

Can also download from Hugging Face: https://huggingface.co/datasets/phamquiluan/DISE-2021

Or from Zenodo: https://zenodo.org/records/12570649

Reproducibility and Evaluation Code

Check the reproduce.ipynb file

Citation

L. Pham, H. Hoang, X.T. Mai, T. A. Tran, "Adaptive Radial Projection on Fourier Magnitude Spectrum for Document Image Skew Estimation", ICIP, 2022.

@inproceedings{pham2021dise,
  title={Adaptive Radial Projection on Fourier Magnitude Spectrum for Document Image Skew Estimation},
  author={Luan Pham, Hao Hoang, Toan Mai, and Tuan Anh Tran},
  booktitle={2022 29th International Conference on Image Processing (ICIP)},
  year={2022},
  organization={IEEE}
}

Metadata

Release files for jdeskew 0.4.2

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for jdeskew 0.4.2
File	Size	Uploaded
jdeskew-0.4.2.tar.gz	845.7 kB	Oct 5, 2026	Details

Built distribution (wheel)

Table of built distributions (wheels) for jdeskew 0.4.2
File	Interpreter	ABI	Platform	Reset
jdeskew-0.4.2-py3-none-any.whl 7.5 kB Oct 5, 2026	Python 3	none	any	Details

Total release size: 853.3 kB

Release files / jdeskew-0.4.2.tar.gz

Download URL	jdeskew-0.4.2.tar.gz
Size	845.7 kB
Tags	Source
SHA-256 checksum How to use checksums	`27b18400172b6b81d700304aa67b67256f2433b7714dd0d48173ed39d9a585cd`
BLAKE2b-256 checksum How to use checksums	`0363f8048f65d813eb3775543d88af37ce4fa861b2cda14febac5abfa49726cd`
Upload date	Oct 5, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/6.2.0 CPython/3.9.25`

Release files / jdeskew-0.4.2-py3-none-any.whl

Download URL	jdeskew-0.4.2-py3-none-any.whl
Size	7.5 kB
Tags	Python 3
SHA-256 checksum How to use checksums	`6d6edf574084ae00244b1fd0d3026124d73a65307300b8e3f2333e677a8b4843`
BLAKE2b-256 checksum How to use checksums	`bbd40b01d016605e484f394140a84b888c5dec2528d32312b11b1dbb136dcb5d`
Upload date	Oct 5, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/6.2.0 CPython/3.9.25`