如何提高 OCR 车牌上数字的可见性

Question

I am using easyocr methods to recognize the text on the license plate but the results are not good.

I have developed deep learning model which detects license plates and crops them.

After that, I just want to detect the numbers on the license plate. What methods can be used to improve the cropped image?

Output from Easy ocr - ['KA51', 'AE-7007']

Output from Easy ocr - ['KLOB BK.7779']

Code:

import easyocr
import string
reader = easyocr.Reader(['en']) 
img = "/mydrive/yolor/cropped/318_0.jpg"

#ALLOWED_LIST = string.ascii_uppercase+string.digits
#characters = reader.readtext(img, detail=0, allowlist=ALLOWED_LIST )
characters = reader.readtext(img, detail=0)
print(characters)}

Answer 1

You can use Pillow to adjust brightness, sharpness and/or contrast. Here is the documentation for the image enhance functions: https://pillow.readthedocs.io/en/stable/reference/ImageEnhance.html

You will have to find something that works well enough on all pictures. That's why I would recommend to do something like a gridsearch and evaluate it on your dataset to find the best settings for the pre processing.