How to extract decimal in image with Pytesseract

Question

Above is the image ,I have tried everything I could get from SO or google ,nothing seems to work. I can not get the exact value in image , I should get 2.10 , Instead it always get 210.

And it is not limited to this image only any image which have a decimal before number 1 tesseract ignores the decimal value.

 def returnAllowedAmount(self,imgpath):
        th = 127
        max_val = 255
        img = cv2.imread(imgpath,0) #Load Image in Memory
        img = cv2.resize(img, None, fx=2.5, fy=2.5, interpolation=cv2.INTER_CUBIC) #rescale Image
        img = cv2.medianBlur(img, 1)
        ret , img = cv2.threshold(img,th,max_val,cv2.THRESH_TOZERO)
        self.showImage(img)

        returnData = pytesseract.image_to_string(img,lang='eng',config='-psm 13 ' )
        returnData = ''.join(p for p in returnData if p.isnumeric() or p == ".") # REMOVE $ SIGN

First thing I would try is to add white border ~20pix around your image. — Dmitrii Z., Aug 13 '19 at 15:12
Unless your attached image is not actually original one, but rather cropped one - letters are very close to the border & this usually causes tesseract mistakes. — Dmitrii Z., Aug 13 '19 at 15:16
Yes , I am cropping image and reading it ,It is working fine with other numbers , but if a a number is followed by a decimal ( . ) , decimal is not read. — PankajKushwaha, Aug 13 '19 at 15:17
Let us [continue this discussion in chat](https://chat.stackoverflow.com/rooms/197884/discussion-between-pankajkushwaha-and-dmitrii-z). — PankajKushwaha, Aug 13 '19 at 15:19

nathancy · Accepted Answer · 2019-08-13T22:49:10.137

Before throwing the image into Pytesseract, some preprocessing to clean/smooth the image helps. Here's a simple approach

Convert image to grayscale and enlarge image
Threshold
Perform morphological operations to clean image
Invert image

First we convert the image to grayscale, resize using the imutils library then threshold to obtain a binary image

Now we perform morphological transformations to smooth the image

Now we invert the image for Pytesseract and add a Gaussian blur

We use the --psm 10 config flag since we want to treat the image as a single character. Here's some additional configuration flags that could be useful

Results

$2.10

After filtering

2.10

import cv2
import pytesseract
import imutils

pytesseract.pytesseract.tesseract_cmd = r"C:\Program Files\Tesseract-OCR\tesseract.exe"

image = cv2.imread('1.png',0)
image = imutils.resize(image, width=300)
thresh = cv2.threshold(image, 150, 255, cv2.THRESH_BINARY_INV)[1]

kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3,3))
close = cv2.morphologyEx(thresh, cv2.MORPH_CLOSE, kernel)

result = 255 - close 
result = cv2.GaussianBlur(result, (5,5), 0)

data = pytesseract.image_to_string(result, lang='eng',config='--psm 10 ')
processed_data = ''.join(char for char in data if char.isnumeric() or char == '.')
print(data)
print(processed_data)

cv2.imshow('thresh', thresh)
cv2.imshow('close', close)
cv2.imshow('result', result)
cv2.waitKey()

Thank You. it worked , I had to change PSM to 12 , since 10 was not returning result. — PankajKushwaha, Aug 14 '19 at 09:29
Do you invert because of the closing? Would not inverting and using opening instead yield the same result? — Joe, Dec 29 '20 at 17:13
@Joe you could do that as well but from my experience that method removes pixels and details from the image making it harder to OCR — nathancy, Feb 17 '21 at 00:55

score 1 · Answer 2 · answered Dec 30 '20 at 19:00

I was able to increase the number of correct decimals by using the methods mentioned in the other answers. Yet, a small share of the decimals were not recognized correctly.

The solution I found was to change the language setting for pytesseract.

I was using a non-English setting, but changing the config to lang='eng' fixed all remaining issues.

Not sure what the reason is, but with the new LSTM engine for Tesseract, the training data is probably mostly English.

score 0 · Answer 3 · answered Jan 16 '20 at 05:49

Sometimes tesseract is oddly sensitive to image size. You can often get better results by scaling your image.

I scaled your image by a factor of 2 and I got good results.

import cv2
import pytesseract

# if windows
# pytesseract.pytesseract.tesseract_cmd = 'C:\\Program Files (x86)\\Tesseract-OCR\\tesseract.exe'

img = cv2.imread('twoten.png', 0)
img = cv2.resize(img, (0,0), fx=2, fy=2)

config = ("--psm 12")

data = pytesseract.image_to_string(img, lang='eng', config = config)

print(data)

which gave this in a console:

$2.10

How to extract decimal in image with Pytesseract

3 Answers3

Linked