OCR: Digitise Physical Text Documents with Ease

Easily convert scanned or photographed documents into editable, machine-readable text using Optical Character Recognition (OCR). From this: ] To This: Method 1: Using AWS Textract API This method leverages Amazon's powerful Textract OCR service to extract text from images. It works well for printed text, especially on structured documents. Features: Batch process multiple files Automatically saves extracted content to .txt files Supports line-by-line text extraction Easy to extend and integrate Sample code: Method 2: Pytesseract Tools We Use Python OpenCV – For image processing Pytesseract – A Python wrapper for Google’s Tesseract-OCR Engine Pillow (PIL) – For additional image support CSV – To export results Before diving into the code, make sure you have Tesseract installed and properly linked in your script. For macOS users: pytesseract.pytesseract.tesseract_cmd = r'/opt/homebrew/bin/tesseract' Step 1: Image Preprocessing Pipeline OCR engines work best when the input text is clear, contrast-rich, and isolated from background noise. Here's the sequence of transformations applied to each image: Grayscale Conversion Converting the image to grayscale helps reduce complexity and makes it easier to isolate text. Sharpening We apply a Laplacian filter to enhance edges and make the text pop. Inversion OCR engines often perform better when text is black on a white background. Thresholding We binarise the image using Otsu’s method, separating text from background. Denoising Removing small artifacts with a median blur filter. Font Smoothing (Optional) If text appears too thin or thick, you can apply morphological operations (dilation/erosion). Step 2: OCR Text Extraction Once the image is processed, we convert it back to RGB (required by Tesseract) and extract the text. Step 3: Save to CSV Finally, we store the extracted text in a CSV file for later use. Sample code: What Next? Once your text is extracted, check out my companion repository: https://github.com/TheOxfordDeveloper/Parsing-unstructured-data.git It shows how to clean and convert raw OCR text into structured, tabular formats, like this:

Apr 29, 2025 - 12:00

OCR: Digitise Physical Text Documents with Ease

Easily convert scanned or photographed documents into editable, machine-readable text using Optical Character Recognition (OCR).

From this:

]

To This:

Method 1: Using AWS Textract API

This method leverages Amazon's powerful Textract OCR service to extract text from images. It works well for printed text, especially on structured documents.

Features:

Batch process multiple files
Automatically saves extracted content to .txt files
Supports line-by-line text extraction
Easy to extend and integrate

Sample code:

Method 2: Pytesseract

Tools We Use

Python
OpenCV – For image processing
Pytesseract – A Python wrapper for Google’s Tesseract-OCR Engine
Pillow (PIL) – For additional image support
CSV – To export results

Before diving into the code, make sure you have Tesseract installed and properly linked in your script. For macOS users:

pytesseract.pytesseract.tesseract_cmd = r'/opt/homebrew/bin/tesseract'

Step 1: Image Preprocessing Pipeline

OCR engines work best when the input text is clear, contrast-rich, and isolated from background noise. Here's the sequence of transformations applied to each image:

Grayscale Conversion
Converting the image to grayscale helps reduce complexity and makes it easier to isolate text.
Sharpening
We apply a Laplacian filter to enhance edges and make the text pop.
Inversion
OCR engines often perform better when text is black on a white background.
Thresholding
We binarise the image using Otsu’s method, separating text from background.
Denoising
Removing small artifacts with a median blur filter.
Font Smoothing (Optional)
If text appears too thin or thick, you can apply morphological operations (dilation/erosion).