I’m working on a license plate detection project which is mostly for educational purposes.
I have gone through some process to locate the license plate and applied perspective transform upon the points of the plate rectangle.

Current issue: I can’t seem to find a proper way to segment characters.
I have tried morphological operations but it sometimes ends up with some extra contours that I can’t distinguish them from actual characters.

Usually there is a segment on the left of the plate or the lines on the right both of which I have marked on the image below:

Code:

mask = cv2.adaptiveThreshold(
    ex_plate, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY_INV, 17, 5
)

mask = cv2.morphologyEx(mask, cv2.MORPH_OPEN, np.ones((3, 3)))

marker = cv2.erode(mask, cv2.getStructuringElement(cv2.MORPH_RECT, (4, 20)))

obr = open_by_reconstruction(marker, mask)

obr = clear_border(obr)

obr = cv2.morphologyEx(obr, cv2.MORPH_CLOSE, np.ones((3, 3)))

obr = hole_filler(obr)

These plates are labeled with correct plate number and I wanted to segment the characters so that I can train a model to identify them.
And this will be the pipeline for each image before reaching the model.