How to force tesseract to read left to right, Python

user_11891146 0
Viewed 902

I have a barchart which I would like to extract the data points from. enter image description here

However, when tesseract reads the image, it is reading left to right AND top to bottom. From my output, you can see that bars with the same height are read in the left to right order:

60.8
58.8 58.8 
58.1
56.9 56.8
54.6 547
51.8 52.2
51:3 
48.7

Jul 2019 Oct 2019 Jan 2020 Apr 2020

I do not want bars of the same height to be read at the same time. instead, I want tesseract to read only left to right (instead of the additional top to bottom it seems to be doing).

I read in another post that the image should be transposed to achieve this, but if I do that, tesseract seems to have trouble reading the transposed image.

Any insight you have would be helpful. Thanks

from PIL import Image, ImageEnhance, ImageFilter
#from pytesseract import image_to_string
import pytesseract
import cv2


pytesseract.pytesseract.tesseract_cmd = r"C:\Program Files\Tesseract-OCR\tesseract.exe"
from pytesseract import image_to_string
im = Image.open(r'C:\Users\Root\im.png')
print(im)

#Resizing and Transposing
new_size = tuple(6*y for y in im.size)
im = im.resize(new_size, Image.ANTIALIAS)
im  = im.transpose(Image.ROTATE_90)
im.save(r'C:\Users\Root\test.png', 'PNG')

#Grayscale for enhanced reading quality
im = cv2.imread(r'C:\Users\Root\test.png')
gray = cv2.cvtColor(im, cv2.COLOR_BGR2GRAY)

print(image_to_string(gray, lang='eng'))
tesseract python
1 Answers

This is unfortunately not the most portable solution, but if you are going to have a lot of bar graphs that look like this, you can slice the image into pieces for each bar, left to right, and run the OCR on each slice separately:

width, height = im.size
numBars = 12
bottomLabelHeight = 100
leftOffset = 10
rightOffset = 20
barWidth = (width - leftOffset - rightOffset) // numBars
for i in range(leftOffset + barWidth, width - rightOffset, barWidth):
    left = i - barWidth
    right = i
    top = 0
    bottom = height - bottomLabelHeight
    bar = im.crop((left, top, right, bottom)).convert('L') # Convert to grayscale
    # bar.show() # uncomment to show bar for testing

    # Need digits whitelist for OCR to work properly
    value = image_to_string(bar, lang='eng', 
        config='--psm 10 --oem 3 -c tessedit_char_whitelist=.0123456789')
    print(value)

    # Read label for bar
    top = height - bottomLabelHeight
    bottom = height
    label = im.crop((left, top, right, bottom)).convert('L')
    print(image_to_string(label, lang='eng'))

Output:

60.8

58.1
Jul 2019
58.8

56.9

51.8
Oct 2019
54.6

56.8

58.8
Jan 2020
54.2

51.3

52.2
Apr 2020
48.7
user_11749660
Related