09/06/2015 at 15:38 •
In this project, we use GOCR to recognize the text on an image.
But it's not perfect. Some words are not corrects, some space are created, some are forget.
So, to correct the text for the speech synthesis, I see 2 options:
- Use a free (as in free beer) spell checker.
- Develop a simple algorithm (merge words to avoid fake spaces and distance algorithm to correct the sentence).
I choose the first solution and I use After The Deadline API. The script will be available on the git repository.
04/24/2015 at 15:31 •
For very visually deficient people, the Open Video Magnifier offers an OCR.
The software makes some steps:
- Get the image:
- Clusterize the text in some groups & give a logical order:
A lot of clustering algorithm without any order exists already. To give an order to the clusters can be a difficulty :).
Example of a very basic clustering algorithm (kmeans):
I will try some algorithm with order later.
- Each group is processed by an open source OCR: GOCR give pretty good results
- Correct the OCR output. Later.
- Give the text to a vocal synthesis. (espeak, festival, etc?)