OCR\OCV RECOGNITION OF UN-EVEN SIZED CHARACTERS

[dheeraj007]

Hello again,

          As a part of exploring vision assitant i tried to read the characters in a region through OCR\OCV recognition through training it.But when trying to read un equal sized character (same characters as before) it is not able to recognize the character.Is there any constraint that OCR\OCV recognition in vision assistant could read characters of equal sizes with equal spacing?? Can u find me a way out...

                                                                           Thank you 🙂

[BruceAmmons]

OCR works on characters that are within about 20% of the original trained size (not sure about exact percentage value).  For characters outside that size range, you could train another instance of the character in a larger size.  You could end up with several instances, each one a different size.  I would probably train instances about 10 or 20 percent larger in each step.

Bruce

[dheeraj007]

Then what about the spacing factor between the characters? I tried to adjust the OCR/OCV setup settings like BOUNDING RECTANGULAR WIDTH  and HEIGHT but it resulted in recognizing the one character as 2  blocks.Is there any solution amicable for this problem? or should i take equally spaced characters.....??

                                                                            thank you...:)

[BruceAmmons]

A lot of the settings are minimum or maximum settings, such as minimum spacing, maximum width, maximum height.  By adjusting them appropriately you should be able to accomodate all the sizes.

If these are strange characters, such as dot matrix, that require fine tuning of the intercharacter spacing, it may not be possible to handle multiple sizes.  Posting some sample images would help the most.

Bruce

[dheeraj007]

Thanks you Mr.Bruce Ammons for your reply.As u intimated i am uploading some of the number plate images with which i am working on.I am also facing an additional problem of setting the ROI to read the characters on the number plate because the number plate in the given below images can be somewhat upward or sideward when compared to one another.So if I set a ROI for one image for OCR\OCV it does not hold good for OCR\OCV of other two images.It would be great help for me if you could specify any method so that the ROI of the OCR\OCV automatically adjusts itself onto the number plate in the image.....

                                                                          Thank you....:)