Your project is very interesting. I am curious, though, as to what preprocessing (if any) you tried using with your nearest neighbor classifier? The example you show on page 11 of your Final Presentation seems like it could be made easier to recognize for NN, if the character were simply centered within the image (calculate the centroid of non-background pixels, and shift).
I am currently (very successfully) localizing black squares with white arrows by OpenCV (two haar cascades - one for arrows down, one for arrows right).
I have my own routines to izolate every character from input grid and I am currently deciding how to recognize handwritten characters inside.
I think that OpenCV's Haar cascades are not good for this problem.
So I was looking for another open-source recognition systems for "handwritten characters" and I found NistOCR.
It should be usable for "our" problem.
It would be nice to keep in touch together and try to figure out it together :).
I should solve this recognition till june or july ;-)
I found that you have done some research in 2008 with k-NN - can you send me presentation (link is not working now) or another results? Do you have it working nicely?
3 comments:
Your project is very interesting. I am curious, though, as to what preprocessing (if any) you tried using with your nearest neighbor classifier? The example you show on page 11 of your Final Presentation seems like it could be made easier to recognize for NN, if the character were simply centered within the image (calculate the centroid of non-background pixels, and shift).
Hello, I am currently working on similar (bachelor degree) project.
Look at my input.
I am currently (very successfully) localizing black squares with white arrows by OpenCV (two haar cascades - one for arrows down, one for arrows right).
I have my own routines to izolate every character from input grid and I am currently deciding how to recognize handwritten characters inside.
I think that OpenCV's Haar cascades are not good for this problem.
So I was looking for another open-source recognition systems for "handwritten characters" and I found NistOCR.
It should be usable for "our" problem.
It would be nice to keep in touch together and try to figure out it together :).
I should solve this recognition till june or july ;-)
I found that you have done some research in 2008 with k-NN - can you send me presentation (link is not working now) or another results? Do you have it working nicely?
Have a nice day!
可愛的小天使~~~支持你~~~.................................................................
Post a Comment