whoisltd / ocr-id-card

extract information values in id card, citizen identity card, passport, ...

OCR-ID-Card

extract information values in id card, citizen identity card, passport, ...

Task

Corner detection task

Detect 4 corners of the id card, passport,… then crop the image. That makes the word detection task easier (reduce noise, variances, …) In this task, using SSD MobileNet v2 to detect corners. SSD has fast inference speed, MobileNet v2 uses low memory, operations but still has good accuracy

Text detection task

Detect some information likely id, name, birthday, … to cropping then use that to extract information task. In this task, using SSD like detect corner task but using Resnet FPN for feature extractor

Extract information task

Extract, get word from the image was cropped from word detection task. Using vietOCR to extract information on images. VietOCR is a pre-trained Vietnamese model that was trained on 10m images. GitHub - pbcquoc/vietocr: Transformer OCR

Run Locally (python3.9.7)

Clone the project

  git clone https://github.com/whoisltd/OCR-ID-Card.git

Go to the project directory

  cd OCR-ID-Card

Install packages

  pip install -r requirements.txt

Start the server

  python3 main.py

Updating ...