Issue recognizing text in image with pytesseract python module

Question

I have the attached an image with 300 DPI. I am using the code below to extract text but I am getting no text. Anyone know the issue?

finalImg = Image.open('withdpi.jpg') text = pytesseract.image_to_string(finalImg)

image to extract text from

Answer 1

Lets observe what is your code doing.

---

• We need to see what part of the text is localized and detected.

• For understanding the code behavior we will use image_to_data function.

• image_to_data will show what part of the image is detected.

• # Open the image and convert it to the gray-scale
  finalImg = Image.open('hP5Pt.jpg').convert('L')
  
  # Initialize ImageDraw class for displaying the detected rectangle in the image
  finalImgDraw = ImageDraw.Draw(finalImg)
  
  # OCR detection
  d = pytesseract.image_to_data(finalImg, output_type=pytesseract.Output.DICT)
  
  # Get ROI part from the detection
  n_boxes = len(d['level'])
  
  # For each detected part
  for i in range(n_boxes):
  
      # Get the localized region
      (x, y, w, h) = (d['left'][i], d['top'][i], d['width'][i], d['height'][i])
  
      # Initialize shape for displaying the current localized region
      shape = [(x, y), (w, h)]
  
      # Draw the region
      finalImgDraw.rectangle(shape, outline="red")
  
      # Display
      finalImg.show()
  
      # OCR "psm 6: Assume a single uniform block of text."
      txt = pytesseract.image_to_string(cropped, config="--psm 6")
  
      # Result
      print(txt)
  

Result:

• i
  I
  

  ```

• 

• So the result is the image itself displays nothing is detected. The code is not-functional. The output does not display the desired result.

• There might be various reasons.

• Here are some facts of the input image:

  • Binary image.

  • Big rectangle artifact.

  • Text is a little bit dilated.

---

• We can't know whether the image requires pre-processing without testing.

• We are sure about the big-black-rectangle is an artifact. We need to remove the artifact. One solution is selecting part of the image.

• To select the part of image, we need to use crop and some trial-and-error to find the roi.

  • If we the image as two pieces in terms of height. We don't want the other artifact containing half.

  • From the first glance, we want (0 -> height/2). If you play with the values you can see that the exact text location is between (height/6 -> height/4)

• Result will be:

• $1,582
  

• 

• Code:

---

• # Open the image and convert it to the gray-scale
  finalImg = Image.open('hP5Pt.jpg').convert('L')
  
  # Get height and width of the image
  w, h = finalImg.size
  
  # Get part of the desired text
  finalImg = finalImg.crop((0, int(h/6), w, int(h/4)))
  
  # Initialize ImageDraw class for displaying the detected rectangle in the image
  finalImgDraw = ImageDraw.Draw(finalImg)
  
  # OCR detection
  d = pytesseract.image_to_data(finalImg, output_type=pytesseract.Output.DICT)
  
  # Get ROI part from the detection
  n_boxes = len(d['level'])
  
  # For each detected part
  for i in range(n_boxes):
  
      # Get the localized region
      (x, y, w, h) = (d['left'][i], d['top'][i], d['width'][i], d['height'][i])
  
      # Initialize shape for displaying the current localized region
      shape = [(x, y), (w, h)]
  
      # Draw the region
      finalImgDraw.rectangle(shape, outline="red")
  
      # Display
      finalImg.show()
  
      # OCR "psm 6: Assume a single uniform block of text."
      txt = pytesseract.image_to_string(cropped, config="--psm 6")
  
      # Result
      print(txt)
  
  

If you can't get the same solution as mine, you need to check your pytesseract version, using:

print(pytesseract.get_tesseract_version())

For me the result is 4.1.1