[FIXED] Is there any way to read Digital Number in the oven using pytesseract OCR?

Issue

I'm doing small project using tesseract OCR. What I want to read is Digital Number which is generated by oven. I pre-treat the image using openCV but, tesseract can't read the image correct. eg. 194 as 794.. let me know is there any way to deal with this. thanks.

the Image which I want to read is shown below

import cv2
import numpy

img_color = cv2.imread('20190509_103247.jpg', cv2.IMREAD_COLOR)

dst = img_color.copy()
roi = img_color[1600:1800,600:1100]
dst[0:200,0:500]=roi
blur = cv2.GaussianBlur(roi,(5,5),0)
gray_dst = cv2.cvtColor(blur, cv2.COLOR_BGR2GRAY)
ret, thr = cv2.threshold(gray_dst, 70, 255,cv2.THRESH_BINARY)
canny = cv2.Canny(roi,100,255)
sobel = cv2.Sobel(gray_dst,cv2.CV_8U,1,0,3)
laplacian = cv2.Laplacian(gray_dst,cv2.CV_8U,ksize=3)
rev = cv2.bitwise_not(canny)

# blur = cv2.GaussianBlur(roi,(5,5),0)

# stencil = numpy.zeros(rev.shape).astype(rev.dtype)
# _, contours, _ = cv2.findContours(rev, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
# color = [255, 255, 255]
# cv2.fillPoly(stencil, contours, color)
# result = cv2.bitwise_and(rev, stencil)

cv2.namedWindow('Show Image')

cv2.imshow('Show Image', rev)

cv2.waitKey(0)

cv2.imwrite('savedimage.jpg', rev)



cv2.destroyAllWindows()

Solution

You can create training data that will work for your font / glyphs to improve how the numeric display gets transformed to the correct digits.

References:

• https://github.com/tesseract-ocr/tesseract/wiki/TrainingTesseract-4.00#creating-training-data
• https://github.com/DevashishPrasad/LCD-OCR

Answered By - dnozay