In this project, I am trying to detect Braille dot groups on a piece of paper and convert it back to original text. I use Template Matching and lookup each letter using its template. I tried working with one template from a certain picture with specific conditions, but it didn’t work for the same paper with different conditions. That made me think that I should rather focus on finding dots on one picture for each template.

In this example, the program detects standard Braille alphabet only, but without abbreviations and shortcuts ( things like “ing”, “he/him”, “th”, “st”, etc). Here’s the result of the experiment:

I started with an initial table of letters and their number of occurrences (ignoring abbreviations, so if there’s an abbreviation for “st” it just counts it as “s” and “t”) as follows:

Without P, Q, Z because they weren’t available in the textbook I picked. Initial

Then I counted the number of occurrences after removing abbreviations. I used this website (https://www.brailletranslator.org/) that converts text to Braille and changes the text to the standard form, so it eliminates any abbreviations and leaves only regular letters

Initial (Without abbreviations)

And these are the results

Evaluation:

Number of correct detected letters: 94
Total number of letters: 139

Success rate of 67.6%

This project was my final for the Computer Vision COM322 class at Connecticut College with Prof. Ozgur Izmirli.