How to perform Optical Character Recognition (OCR)

Steps to detect and OCR with OpenCV and Tesseract:

1. Take and save an image of the item (card/mailpiece/receipt) to be OCR'ed

• This step is not included in this repository, but is as simple as clicking a picture using a camera and saving it to the local machine

  

2. Preprocessing the input image using OpenCV:

• Apply edge detection to reveal the outline (detect the four corners) of the item in the image against the background

  • This step works well if there is sufficient contrast between the background and foreground

    
    

• Detect contours in the edge map

  

• Loop over all contours and find the largest contour with four vertices (corners)

  

• Apply a perspective transform to obtain a top-down, bird’s-eye view of the item (required to improve OCR accuracy)

  
  
  

3. OCR the text on the top-down transform of the item using Tesseract

• Extract data using Python regular expressions

• Display the results on our terminal and save to a text file

  Nihar Garg, EIT, LEED Green Associate
  Project Engineer
  ngarg@princetonengineering Services
  609-216-7499
  101 Morgan Lane STE 205, Plc
  609-452-9700
  609-751-9545

A special thank you to pyimagesearch.