OCR Segmentation and skeleton

[Gary_Guan]

HI,GuysI have a trouble on OCR, anyone can give me a help or some advices? Thanks

The detail is recognizing character from a label, After some processing  (see pictures),the correct recognition rate is still unsatisfactory . I try to segment and extract the skeleton of every character. Combination of the above operations,the result may be better. I knew there are special function VI(Segmentation.vi, Skeleton.vi), also have tried them,but failed. Who can tell me how to using them or other method.

Thanks

Best regards

    --->    ---->

[Andrey_Dmitriev]

Hi, Gary,

Is simple thresholding, embedded in NI OCR not good enough for you?

I don't think that Segmentation / Skeleton is right way for you. If you want to perform segmentation in your own way, then you should do it in assumption that your segments (letters in this case) and they have predefined sizes/spaces. Use this prior knowledge. Your image looks pretty uniform, so you can do intitial threshold, then analyse particles, then remove particles which are too small, probably perform threshold again with higher value in areas where particles too large for better separating and so on... Do it iterative. Sorry, i'm not guru in segmentation theory.

with best regards, 

Andrey.

[Gary_Guan]

Hi, Andrey

Thanks for you reply.I will have a try as you said.

And could you tell me you what is the tool in the picture. is ''NI OCR Training interface"?

best regards,

Gary

[Andrey_Dmitriev · Accepted solution]

---

Gary Guan wrote:

And could you tell me you what is the tool in the picture. is ''NI OCR Training interface"?

---

This is Vision Assistant 2009. OCR Training Interface is included into Vision Assistant as well.

Andrey.