Hello I'm trying to ocr on a special image.
I've done adaptive threshold , polartocartesian transformation.
this is my original image
this is adaptive threshold
this is polartocartesian transformation
what do you think is this image good enough for OCR or do I have to optimize my preprocessing steps ?
It looks like using the polartocartesian transformation is destroying the characters (look at the "M", "E" before and after...). I think it will be a better idea to think about some rotations mechanism using imrotate.
Any way, I'd try the following additional pre-proc:
1. you might want to perform some morphological operations on the image after the adaptive threshold.
2. I would also try labeling the image (bwlabel) and removing too large objects (most of the background noise in your image is connected so it will be a huge object - in terms of pixel counting)
Have fun!
The main issue here is the directional lighting which illuminates mainly one side of the characters, so that a part of them disappears in the background. If you rotate the part, the appearance of the letters will change; this is bad for recognition. You may also face fragmentation of the characters (K for example).
I don't like this adaptive thresholding at all, as it reveals a lot of the texture and could very well cause the characters to touch surrounding blobs and be impossible to segment out. You should increase the adaptiveness scale or use global thresholding.
If possible, improve illumination using a diffuse source and let the background disappear.
No preprocessing step will restore what adverse illumination has lost.
Signs on your images look quite well. I think you should optimize your preprocessing step and use more complex filters. There is very interesting answer of tbirdal - How to remove the noise without destroying the main edge? See on Mazda logo before and after filtration in above-mentioned answer.
Rodrane, this is very tough. You should either adjust your lighting or adapt your threhsolding algorithm. Also apply some smoothing (possibly edge preserving) before doing the binarization and polar transformation - first do the polar transform, then binarize. For a solution of a similar problem, please see this SE answer.
