For myself I wrote some dirty script with pytesseract
and few techniques from opencv
library. You can choose different params here and view results.
For example, I have image with name softserve.png
:

Suppose you have ocr.py
with following code:
# import the necessary packages
import argparse
import cv2
import numpy as np
import os
from PIL import Image
import pytesseract
# construct the argument parse and parse the arguments
ap = argparse.ArgumentParser()
ap.add_argument("-i", "--image", required=True,
help="path to input image to be OCR'd")
ap.add_argument("-b", "--blur", type=str, default=None,
help="type of preprocessing to be done")
ap.add_argument("-t", "--thresh", type=str, default=None,
help="type of preprocessing to be done")
ap.add_argument("-r", "--resize", type=float, default=1.0,
help="type of preprocessing to be done")
ap.add_argument("-m", "--morph", type=str, default=None,
help="type of preprocessing to be done")
args = vars(ap.parse_args())
# load the example image and convert it to grayscale
image = cv2.imread(args["image"])
# Resize to 2x
if args["resize"] != 1:
image = cv2.resize(image, None,
fx=args["resize"], fy=args["resize"],
interpolation=cv2.INTER_CUBIC)
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
kernel = np.ones((5, 5), np.uint8)
# make a check to see if median blurring should be done to remove
# noise
if args["blur"] == "medianblur":
gray = cv2.medianBlur(gray, 3)
if args["blur"] == "avgblur":
gray = cv2.blur(gray, (5, 5))
if args["blur"] == "gaussblur":
gray = cv2.GaussianBlur(gray, (5, 5), 0)
if args["blur"] == "medianblur":
gray = cv2.medianBlur(gray, 3)
if args["blur"] == "filter":
gray = cv2.bilateralFilter(gray, 9, 75, 75)
if args["blur"] == "filter2d":
smoothed = cv2.filter2D(gray, -1, kernel)
# check to see if we should apply thresholding to preprocess the
# image
if args["thresh"] == "thresh":
gray = cv2.threshold(gray, 0, 255,
cv2.THRESH_BINARY | cv2.THRESH_OTSU)[1]
if args["thresh"] == "thresh1":
gray = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY)[1]
if args["thresh"] == "thresh2":
gray = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY_INV)[1]
if args["thresh"] == "thresh3":
gray = cv2.threshold(gray, 0, 255, cv2.THRESH_TRUNC)[1]
if args["thresh"] == "thresh4":
gray = cv2.threshold(gray, 0, 255, cv2.THRESH_TOZERO)[1]
if args["thresh"] == "thresh5":
gray = cv2.threshold(gray, 0, 255, cv2.THRESH_TOZERO_INV)[1]
if args["thresh"] == "thresh6":
gray = cv2.adaptiveThreshold(gray, 255,
cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
cv2.THRESH_BINARY, 115, 1)
if args["thresh"] == "thresh7":
gray = cv2.adaptiveThreshold(gray, 255, cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY, 115, 1)
if args["thresh"] == "thresh8":
gray = cv2.adaptiveThreshold(gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
cv2.THRESH_BINARY, 11, 2)
if args["thresh"] == "thresh9":
gray = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)[1]
if args["thresh"] == "thresh10":
# gray = cv2.GaussianBlur(gray, (5, 5), 0)
gray = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)[1]
if args["morph"] == "erosion":
gray = cv2.erode(gray, kernel, iterations=1)
if args["morph"] == "dilation":
gray = cv2.dilate(gray, kernel, iterations=1)
if args["morph"] == "opening":
gray = cv2.morphologyEx(gray, cv2.MORPH_OPEN, kernel)
if args["morph"] == "closing":
gray = cv2.morphologyEx(gray, cv2.MORPH_CLOSE, kernel)
# write the grayscale image to disk as a temporary file so we can
# apply OCR to it
filename = "{}.png".format(os.getpid())
cv2.imwrite(filename, gray)
# load the image as a PIL/Pillow image, apply OCR, and then delete
# the temporary file
text = pytesseract.image_to_string(Image.open(filename))
os.remove(filename)
print(text)
with open("output.py", "w") as text_file:
text_file.write(text)
# show the output images
cv2.imshow("Image", image)
cv2.imshow("Output", gray)
cv2.waitKey(0)
If I simply use usual OCR without anything (such as pytesseract.image_tostring()
):
python3 ocr.py --image softserve.png
I would got this text:
uray ['Amir', 'Barry', 'Chales', ‘Dao']
‘amir’ rss
tee)
print(2)
It's a very bad result, isn't it?
But after playing with resize
and thresh
you can got a more nice output:
python3 ocr.py --image softserve.png --thresh thresh6 --resize 2.675
And see in two opened windows how looks image before OCR:

Output:
names1 = ['Amir', ‘Barry’, ‘Chales', ‘Dao']
if ‘amir' in names1:
@ print(1)
else: «=
@ print(2)
You also can apply morph
and blur
. You can read more about blur, thresholding and morphological transformations from opencv
docs. I am hope, you will find that information useful in your work