I am trying to develop a basic OCR for bangla character using openCV. I have used opencv's built-in CvKNearest object. everything works alright except the accuracy of recognition. I haven't used any pre-calculated feature. I have plan to use some more feature to increase the accuracy. I have made some googling to find which feature is used in Opencv's Knearest object for prediction, but I couldn't get that. could anyone please explain me Which feature vector does opencv's K-Nearest-Neighbor algorithm use to predict in pattern recognition?
Asked
Active
Viewed 505 times
1 Answers
1
If you are looking for what kind of features you need, there are lot .
- Statistical features - ex , normalized density values , zone method, # "on" pixels
- Structural features - ex , # intersections, # holes etc
- Projection histogram features etc
Like wise there are lot. Suggest you to read about few publications regarding OCR first, start from here example

Heshan Sandeepa
- 3,388
- 2
- 35
- 45