Program to extract signatures from documents.

Overview

Extracting Signatures from Bank Checks

Introduction

Ahmed et al. [1] suggest a connected components-based method for segmenting signatures in document images. For this purpose, they train and evaluate their models on Tobacco-800 documents dataset to extract patch-level signatures.

In this project, we have taken their proposal and modified it to work specifically for bank checks, and be able to segment signatures on a stroke level instead of patch level. Both these are inherently more difficult problems.

Bank checks have complex backgrounds which are not known beforehand. This was not the case in Tobacco-800 dataset where the documents have a simple white background. Similarly, extracting stroke-level position of the signature is again a more difficult task compared to patch level location.

Demo

You can watch a demo of the results in the following Youtube video
Demo Video

Datasets

One reason why [1] worked on patch-level was a lack of publicly available datasets with groundtruth at stroke-level. In this project, we have created our own dataset of bank checks with stroke level groundtruth available as binary segmentation masks. We use this dataset for testing.

We divide our training data into two classes: signatures and non-signature. For non-signature class, we took 64 documents from Tobacco-800 [3, 4] to manually generate new images which which didn't contain any signatures or handwritten text. Some of these documents contained logos, as well. For the signature class, we combined all images from the UTSig signature database [2], which contains 8280 Persian language signatures.

Installation

Clone the project and cd to project directory. Then, execute the following commands once to set-up your environment

python3 -m venv venv/
source venv/bin/activate
pip install -r requirements.txt

Then you can run the GUI app with python3 src/extract_gui.py.

The repository comes with trained models in the models/ folder. To train a new model with your own data, look at the train_test.py script.

References

  1. Ahmed, S., Malik, M. I., Liwicki, M., & Dengel, A. (2012, September). Signature segmentation from document images. In 2012 International Conference on Frontiers in Handwriting Recognition (pp. 425-429). IEEE.

  2. Soleimani, A., Fouladi, K., & Araabi, B. N. (2016). UTSig: A Persian offline signature dataset. IET Biometrics, 6(1), 1-8.

  3. Zhu, G., & Doermann, D. (2007, September). Automatic document logo detection. In Ninth International Conference on Document Analysis and Recognition (ICDAR 2007) (Vol. 2, pp. 864-868). IEEE.

  4. Zhu, G., Zheng, Y., Doermann, D., & Jaeger, S. (2007, June). Multi-scale structural saliency for signature detection. In 2007 IEEE Conference on Computer Vision and Pattern Recognition (pp. 1-8). IEEE.

Owner
Muhammad Saif Ullah Khan
Mobile developer and ML engineer.
Muhammad Saif Ullah Khan
一款不需要买代理来减少扫网站目录被封概率的扫描器,适用于中小规格字典。

PoorScanner使用说明书 -工具在不同环境下可能不怎么稳定,如果有什么问题恳请大家反馈。说明书有什么错误的地方也大家欢迎指正。 更新记录 2021.8.23 修复了云函数主程序 gitee上传文件接口写错了的BUG(之前把自己的上传地址写死进去了,没从配置文件里读) 更新了说明书 PoorS

14 Aug 02, 2022
Obsidian tools - a Python package for analysing an Obsidian.md vault

obsidiantools is a Python package for getting structured metadata about your Obsidian.md notes and analysing your vault.

Mark Farragher 153 Jan 04, 2023
Software to help automate collecting crowdsourced annotations using Mechanical Turk.

Video Crowdsourcing Software to help automate collecting crowdsourced annotations using Mechanical Turk. The goal of this project is to enable crowdso

Mike Peven 1 Oct 25, 2021
A simple toolchain for moving Remarkable highlights to Readwise

A simple toolchain for moving Remarkable highlights to Readwise

zach wick 20 Dec 20, 2022
A primitive Python wrapper around the Gromacs tools.

README: GromacsWrapper A primitive Python wrapper around the Gromacs tools. The library is tested with GROMACS 4.6.5, 2018.x, 2019.x, 2020.x, and 2021

Becksteinlab 140 Dec 28, 2022
kawadi is a versatile tool that used as a form of weapon and is used to cut, shape and split wood.

kawadi kawadi (કવાડિ in Gujarati) (Axe in English) is a versatile tool that used as a form of weapon and is used to cut, shape and split wood. kawadi

Jay Vala 2 Jan 10, 2022
Python module and its web equivalent, to hide text within text by manipulating bits

cacherdutexte.github.io This project contains : Python modules (binary and decimal system 6) with a dedicated tkinter program to use it. A web version

2 Sep 04, 2022
A BlackJack simulator in Python to simulate thousands or millions of hands using different strategies.

BlackJack Simulator (in Python) A BlackJack simulator to play any number of hands using different strategies The Rules To keep the code relatively sim

Hamid 4 Jun 24, 2022
This project is a set of programs that I use to create a README.md file.

This project is a set of programs that I use to create a README.md file.

Tom Dörr 223 Dec 24, 2022
Prime Path Generator is a prime path generator used to generate prime paths.

Prime Path Generator is a prime path generator used to generate prime paths.

1 Nov 06, 2021
Python HTTP Agent Parser

Features Fast Detects OS and Browser. Does not aim to be a full featured agent parser Will not turn into django-httpagentparser ;) Usage import ht

Shekhar 213 Dec 06, 2022
Simple profile athena generator for Fortnite Private Servers.

Profile-Athena-Generator A simple profile athena generator for Fortnite Private Servers. This profile athena generrator features: Item variants Get al

Fevers 10 Aug 27, 2022
A thing to simplify listening for PG notifications with asyncpg

asyncpg-listen This library simplifies usage of listen/notify with asyncpg: Handles loss of a connection Simplifies notifications processing from mult

ANNA 18 Dec 23, 2022
Color getter (including method to get random color or complementary color) made out of Python

python-color-getter Color getter (including method to get random color or complementary color) made out of Python Setup pip3 install git+https://githu

Jung Gyu Yoon 2 Sep 17, 2022
A library to easily convert climbing route grades between different grading systems.

pyclimb A library to easily convert climbing route grades between different grading systems. In rock climbing, mountaineering, and other climbing disc

Ilias Antonopoulos 4 Jan 26, 2022
A quick username checker to see if a username is available on a list of assorted websites.

A quick username checker to see if a username is available on a list of assorted websites.

Maddie 4 Jan 04, 2022
Control-Alt-Delete - Help Tux Escape Beastie's Jail!

Control-Alt-Delete Help Tux escape Beastie's jail by completing the following challenges! Challenges Challenge 00: Drinks: Tux needs to drink less. Ch

NDLUG 8 Oct 31, 2021
Simple RGB to HEX game made in python

Simple RGB to HEX game made in python

5 Aug 26, 2022
ecowater-softner is a Python library for collecting information from Ecowater water softeners.

Ecowater Softner ecowater-softner is a Python library for collecting information from Ecowater water softeners. Installation Use the package manager p

6 Dec 08, 2022
Greenery - tools for parsing and manipulating regular expressions

Greenery - tools for parsing and manipulating regular expressions

qntm 242 Dec 15, 2022