ANPR - Automatic number plate recognition
Identify a vehicle license plate using image processing.
The chalenge is identify which region of the image is the plate and after apply an ocr over this region.
The purpose of this sample is not explain how to implement an OCR but how to find the plate for a given image.
Steps to achieve the goal
-
Load the image and convert to grey scale
-
Apply a gaussian filter to remove noise
-
Apply more filters to transform to B&W and remove dark spots
-
Find the edges of relevant regions of the image
-
Apply OCR over these regions and get the plate number and get a collection of text.
-
Iterate over this collection of text to check if this is a car plate or other info like a sticker, advertising or other text in some region of the image.
If you are working with fixed camera you can also check the position (x,y) of each region and avoid unecessary OCR processing since usually the plate is located in the same position in all vehicles.
What you need
- python 2.7 ~ 3.x (used v 3.7.6)
- some python libs (see bellow)
Install python libs
pip3 install sklearn
pip3 install matplotlib
Run
% python3 anpr.py
Expected Result
Original Image
Image converted to greyscale
Noise reduction
Remove small dark spots and connect small bright cracks
Clear objects connected to the label image border
Cropped regions
and...
The Plate !
Testing with OCR
After pushing the image plate region to this free OCR https://www.newocr.com/ received the text: MUIEPSD - \o/
Other techniques
You can try other filters or adjust the parameters to achieve a better accuracy and performance.
