mocr

Meaningful Optical Character Recognition from identity cards with Deep Learning.

Introduction

mocr is a library that can be used to detect meaningful optical characters from identity cards. Code base is pure Python and works with 3.x versions. It has some low level dependencies such as Tesseract. mocr uses a pre-trained east detector with OpenCV and applies it's Deep Learning techniques.

It has a pre-trained east detector inside the module and a custom trained model can be given as a parameter.

Prerequisites

• Tessaract must be installed on your computer before using OCR. Please check installation link for details.
• The other dependencies are listed on requirements.txt and will be installed when you install with pip.

Installation

From source

Install module using pip:

$ pip install mocr

Download the latest mocr library from: https://github.com/verifid/mocr

Install module using pip:

$ pip install -e .

Extract the source distribution and run:

$ python setup.py build
$ python setup.py install

Running Tests

The test suite can be run against a single Python version which requires pip install pytest and optionally pip install pytest-cov (these are included if you have installed dependencies from requirements.testing.txt)

To run the unit tests with a single Python version:

$ py.test -v

to also run code coverage:

$ py.test -v --cov-report html --cov=mocr

To run the unit tests against a set of Python versions:

$ tox

Sample Usage

• text_recognition Initiating the TextRecognizer with identity image and then finding the texts with their frames:

import os
from mocr import TextRecognizer

image_path = os.path.join('tests', 'data/sample_uk_identity_card.png')
east_path = os.path.join('mocr', 'model/frozen_east_text_detection.pb')

text_recognizer = TextRecognizer(image_path, east_path)
(image, _, _) = text_recognizer.load_image()
(resized_image, ratio_height, ratio_width, _, _) = text_recognizer.resize_image(image, 320, 320)
(scores, geometry) = text_recognizer.geometry_score(east_path, resized_image)
boxes = text_recognizer.boxes(scores, geometry)
results = text_recognizer.get_results(boxes, image, ratio_height, ratio_width)

# results: Meaningful texts with bounding boxes

• face_detection:

from mocr import face_detection

image_path = 'YOUR_IDENTITY_IMAGE_PATH'
face_image = face_detection.detect_face(image_path)
# face_image is the byte array detected and cropped image from original image

from mocr import face_detection

video_path = 'YOUR_IDENTITY_VIDEO_PATH'
face_image = face_detection.detect_face_from_video(video_path)
# face_image is the byte array detected and cropped image from original video

CLI

Sample command line usage

• Optical Character Recognition

python -m mocr --image tests/data/sample_uk_identity_card.png --east tests/model/frozen_east_text_detection.pb

• Face detection from image file

python -m mocr --image-face 'tests/data/sample_de_identity_card.jpg'

• Face detection from video file

python -m mocr --video-face 'tests/data/face-demographics-walking.mp4'

Screenshots

Before

After