Tesseract ocr failed to recognize number from number plate images

asked 2019-10-22 04:12:19 -0500

updated 2019-10-22 07:40:23 -0500

supra56 gravatar image

Code:

import cv2
import numpy as np
from PIL import Image
import pytesseract
from scipy import ndimage
from scipy.ndimage import rotate
#from matplotlib import pyplot as plt

import allow_needed_values as anv

img = cv2.imread('path_to_image/1.jpg')
result = cv2.fastNlMeansDenoisingColored(img,None,10,10,7,21)
cv2.imshow("denoise",result)
gray = cv2.cvtColor(result, cv2.COLOR_RGB2GRAY)
cv2.imshow("gray image",gray)

gray = cv2.threshold(gray, 0, 255,cv2.THRESH_BINARY | cv2.THRESH_OTSU)[1]
cv2.imshow("threshold image", gray)
#cv2.waitKey(delay = 0)
#gray = cv2.medianBlur(gray, 3)
img = cv2.medianBlur(gray,3)
cv2.imshow("median blur", img)
#rot = rotate(img, -6, reshape=False)
#rotated= ndimage.rotate(img,-2)
#cv2.imshow("rotate image", rot)
#result = cv2.fastNlMeansDenoisingColored(img,None,20,10,7,21)
rot = rotate(img, -5, reshape=False)
cv2.imshow("rotate", rot)
tessdata_dir_config = "/usr/share/tesseract-ocr/4.00/tessdata/"
text=pytesseract.image_to_string(rot,lang='eng',config=tessdata_dir_config)
carReg = anv.catch_rectify_plate_characters(text)
print(carReg)
cv2.waitKey(delay = 0)

Here i write the code for number recognition from number plate images,but the tessaract failed to recognize number. I have day light and night images. in most of the case ocr failed on night images. also i apply various filter filter for images clearing and also rotate the image angle for some degree.

I upload the various sample images where tesseract ocr failed. image description image description image description image description image description image description

edit retag flag offensive close merge delete

Comments

1

Well i am facing somehow the same problem when trying to read text from various backgrounds. Tesseract is pretty limited(for something which uses a neural network): Its expects black letters on white background with 300 dpi resolution.

So i uses kmeans for background removal(color clustering) and the i resize to 300 dpi before feeding it to tesseract. It improved the results a lot. I am still very unhappy with the quality of tesseract.

holger gravatar imageholger ( 2019-10-22 05:45:40 -0500 )edit