Introduction

Chinese Plate Number Recognition powered by OpenCV and MXNet written in Python 3.

Based on the following Github projects:

• https://github.com/szad670401/end-to-end-for-chinese-plate-recognition

• https://github.com/ibyte2011/end-to-end-for-chinese-plate-recognition

Dependencies

• Python 3

• OpenCV 3.2

• MXNet 0.9.5

Please see notes/opencv.md and notes/mxnet.md for installing OpenCV and MXNet, respectively.

The following combinations were tested successfully:

• Python 3.5 + OpenCV 3.2.1 + MXNet 0.9.5(with GPU) + Ubuntu 16.04.2

• Python 3.6 + OpenCV 3.2.1 + MXNet 0.9.5(without GPU) + Mac OS Sierra (10.12.5)

File Structure

• fonts/ - fonts that are used to generate plate number images.

• images/ - background images.

• models/ - trained MXNet models.

• no_plates/ - noise images which contain no plate numbers.

• plates/ - plate number images for training.

• plates_to_test/ - plate number images for testing.

Usage

1. Generate plate images for training:

python3 ./generate_plates.py

2. Training:

python3 ./train.py

3. Recognition:

python3 ./recognize.py