Streamlit Tesseract OCR 🔎 📄

Streamlit project with Tesseract OCR running on Streamlit Cloud.

App Usage 💻

1. Upload an image with text on it
2. Select the language
3. Select the image preprocessing options (if needed) and check the result in the preview
4. Crop the image to the text area (if needed)
5. Run the OCR and check the result in the text preview
6. Adjust the settings or image preprocessing and run the OCR again (if needed)
7. Download the result as a text file or copy from the text preview

Languages 🌍

Installed languages for Tesseract OCR

🇬🇧 🇪🇸 🇫🇷 🇩🇪 🇮🇹 🇵🇹 🇨🇿 🇵🇱

Status ✔️

Streamlit application is working - 04.06.2024

ToDo 📝

• Change layout of the app
• Change checkboxes to toggle buttons
• Add cropping functionality: https://github.com/turner-anderson/streamlit-cropper
• Add more CSS styling
• Cleanup of python app and repository

Future Ideas 💡

• Use Pillow for image preprocessing instead of OpenCV
  • any advantages?
• Add Ace Editor for text preview
  • any advantages?
• Add other OCR engines and test them
• Add easyocr and test it
  • https://github.com/JaidedAI/EasyOCR
• Try tesserocr instead of pytesseract
  • https://github.com/sirfz/tesserocr
• Add PyMuPDF and test it
  • https://github.com/pymupdf/PyMuPDF
• Add ocrmypdf and test it
  • https://github.com/ocrmypdf/OCRmyPDF
• Add PaddleOCR and test it
  • https://github.com/PaddlePaddle/PaddleOCR
• Add keras-ocr and test it
  • https://github.com/faustomorales/keras-ocr

Libraries 📚

Tesseract

• Tesseract Documentation
  • https://tesseract-ocr.github.io/
• pytesseract Documentation
  • https://github.com/madmaze/pytesseract
• OCR with Tesseract
  • https://nanonets.com/blog/ocr-with-tesseract/

EasyOCR

• https://github.com/JaidedAI/EasyOCR

pdf2image

• https://github.com/Belval/pdf2image

OpenCV

OpenCV is used for image preprocessing before running OCR with Tesseract.

• OpenCV Image Processing Documentation
  • https://docs.opencv.org/4.x/d2/d96/tutorial_py_table_of_contents_imgproc.html
• OpenCV Python Tutorial
  • https://www.geeksforgeeks.org/opencv-python-tutorial/
• OCR in Python Tutorials
  • https://www.youtube.com/watch?v=tQGgGY8mTP0&list=PL2VXyKi-KpYuTAZz__9KVl1jQz74bDG7i

Pillow

• https://pillow.readthedocs.io/en/stable/

Image Preprocessing 🖼️

Grayscale Conversion

import cv2
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)

# or
coefficients = [1,0,0] # Gives blue channel all the weight
# for standard gray conversion, coefficients = [0.114, 0.587, 0.299]
m = np.array(coefficients).reshape((1,3))
blue = cv2.transform(im, m)

Brightness and Contrast

• CLAHE (Contrast Limited Adaptive Histogram Equalization)
• https://www.tutorialspoint.com/how-to-change-the-contrast-and-brightness-of-an-image-using-opencv-in-python
• https://stackoverflow.com/questions/50474302/how-do-i-adjust-brightness-contrast-and-vibrance-with-opencv-python
• https://stackoverflow.com/questions/32609098/how-to-fast-change-image-brightness-with-python-opencv
• https://github.com/milahu/document-photo-auto-threshold
• https://stackoverflow.com/questions/56905592/automatic-contrast-and-brightness-adjustment-of-a-color-photo-of-a-sheet-of-pape
• https://stackoverflow.com/questions/39308030/how-do-i-increase-the-contrast-of-an-image-in-python-opencv
• https://stackoverflow.com/questions/63243202/how-to-auto-adjust-contrast-and-brightness-of-a-scanned-image-with-opencv-python

Image Rotation 🔄

Methods to rotate an image with different libraries.

... with Pillow 🔄

https://pillow.readthedocs.io/en/stable/reference/Image.html#PIL.Image.Image.rotate

from PIL import Image
with Image.open("hopper.jpg") as im:
    # Rotate the image by 60 degrees counter clockwise
    theta = 60
    white = (255,255,255)
    # Angle is in degrees counter clockwise
    im_rotated = im.rotate(angle=theta, resample=Image.Resampling.BICUBIC, expand=1, fillcolor=white)

... with OpenCV 🔄

destructive rotation, loses image data

import cv2
(h, w) = image.shape[:2]
center = (w // 2, h // 2)
M = cv2.getRotationMatrix2D(center, angle, 1)
rotated = cv2.warpAffine(image, M, (w, h))

... with imutils 🔄

non-destructive rotation, keeps image data

import imutils
rotate = imutils.rotate_bound(image, angle)

... with scipy 🔄

destructive or non-destructive rotation, can be chosen py parameter reshape

from scipy.ndimage import rotate as rotate_image
rotated_img1 = rotate_image(input, angle, reshape, mode, cval)