License Plate Detection

This project is done as a Final Year Project required to complete a Bachelor's degree in Electronics Engineering. The code is based on Adrian Rosebrock's (@jrosebr1) PyImageSearch post OpenCV: Automatic License/Number Plate Recognition (ANPR) with Python, edited to be able to detect Malaysian license plates which typically has white texts on black background plate. I recommend reading through his blogpost to get a general idea of how the image processing pipeline works.

The project aims to:

• Automatically detects either Malaysian license plates or other oversea license plates
• Compare Scharr operator, Canny edge detector, and edge-less approach to detect license plates

Prerequisites

The project code utilizes the following tools:

• Docker
• Python v3.8.6
• OpenCV v4.4.0.46
• imutils v0.5.3
• scikit-image v0.17.2
• pytesseract v0.3.6
• Tesseract-OCR v5.0.0 (alpha)

How it works

1. The project has 2 modules which is the driver and the class module named anprdriver.py and anprclass.py respectively. The driver module contains the script which the user should execute through Python while providing the arguments whereas the class module contains the three detection algorithm (Sobel, Canny, and edge-less) and other functions including saving results and workflow debug display.
2. The driver should be executed in a terminal by using Python ($ python3 anprdriver.py ...). The arguments which could be appended are (arguments in bold are required):
   • -i (input) - path to input directory of images
   • -c (clear border) - should the script clear border pixels before performing OCR (default is -1, set to 1 to enable)
   • -p (PSM) - default tesseract page segmentation mode for license plate OCR (default is 7)
   • -d (debug) - should additional visualizations be shown (default is -1, set to 1 to enable)
   • -a (algorithm) - specify which algorithm to use (default is 1, 1 for Scharr, 2 for Canny, 3 for edge-less)
   • -s (save) - whether to save the results or not (default is -1, disable debug and set to 1 to enable)
   • -m (morphology) - specify morphological operation (default is 1, 1 for black hat which is better for black text on white plates, 2 for top hat which is better for the opposite)
3. The repository includes sample images of license plates in malaysian and overseas folder. Results are saved inside result_canny_*, result_sobel_*, and result_edgeless_* folders respectively where * is either malaysian or overseas.

Typical usage

Build the Docker image docker build -t anpr . first then run in interactive mode docker run -it anpr.

The script can then be typically run by typing $ python3 anprdriver.py -i /path/to/images/ in the container terminal. Path to the images can be written as -i /malaysian for example. While only -i argument is necessary, be sure to choose the correct algorithm (-a) and morphological operation (-m) as needed.

You can also include your own source image by modifying the Dockerfile.