Skip to content
/ CEIR Public
forked from eadst/CEIR

Code for Novel Receipt Recognition with Deep Learning Algorithms

Notifications You must be signed in to change notification settings

hadils/CEIR

 
 

Repository files navigation

CEIR

This project is for the SPIE paper - Novel Receipt Recognition with Deep Learning Algorithms. In this paper, we propose an end-to-end novel receipt recognition system for capturing effective information from receipts (CEIR).

CEIR system demo is available at: CEIR Demo

The CEIR has three parts: preprocess, detection, recognition.

Introduction

In the preprocessing method, by converting the image to gray scale and obtaining the gradient with the Sobel operator, the outline of the receipt area is decided by morphological transformations with the elliptic kernel.

In text detection, the modified connectionist text proposal network to execute text detection. The pytorch implementation of detection is based on CTPN.

In text recognition, the convolutional recurrent neural network with the connectionist temporal classification with maximum entropy regularization as a loss function to update the weights in networks and extract the characters from receipt. The pytorch implementation of recognition is based on CRNN and ENESCTC.

We validate our system with the scanned receipts optical character recognition and information extraction (SROIE) database.

Dependency

Python 3.6.3

  1. torch==1.4
  2. torchvision
  3. opencv-python
  4. lmdb

Prediction

  1. Download pre-trained model from Google Drive and put the file under ./detection/output/ folder.

  2. Change the image name to demo.jpg in the CEIR folder.

  • Run python ceir_crop.py for stage 1.
  • Run python ceir_detect.py for stage 2.
  • Run python ceir_recognize.py for stage 3.
  1. The result will be saved in ./result/.

Training

  1. Put dataset in ./dataset/train/image and ./dataset/train/label.

  2. Preprocess parameters can be changed in ./preprocess/crop.py.

  3. In the detection part, the ./detection/config.py is used for configuring. After that, run python train.py in the detection folder.

  4. In recognition, you need to change trainroot and other parameters in train.sh, then run sh train.sh to train.

About

Code for Novel Receipt Recognition with Deep Learning Algorithms

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 99.6%
  • Shell 0.4%