Experiment Three

  • Attempt to create a set of 96 pictures-of-letters to use for Template Matching.
  • Show.
  • Provide.

Explain. Illustrate. Summarize concisely.

Here we endeavor to see if the image of the Riverbank Laboratories 1916 lower case decoder card is suitable for slicing up into individual pictures-of-letters.

The following shows the output from the findContours routine of the Computer Vision library, CV2:

The algorithm is so exquisitely precise that it produces a superabundance of spurious bounding boxes.

Also, the OCR text extracted by CV2 is, not surprisingly, gibberish.

But as mentioned before, the task is inherently simple, the product of printing technology from 1623, and the algorithm for filtering the spurious bounding boxes is refreshingly simple, see ‘remove_duplicate_bounding_boxes’ in the Python source code:



View the Jupiter Notebook: the experimental commands used here, and the results recorded.

View and download the Python source code from Github.

Independently run the program on Google Colab, or create a clone of the program to run on a Colab account of your own.

Notes for this page:

  • First additional note.
Scroll to Top