Machine Vision

cancel
Showing results for 
Search instead for 
Did you mean: 

Hi,Please reply back if you had a solution to the problems listed below using OCR:-

 1) Characters with very small font are not recognized(Font of 4).(Eg :- Press ENTER to submit or ESC to cancel )

 2) How can i read Black Character with Gray Background

    Should i have to train separately for  these?

3) Letters which are trained in an image are not giving the same output when ROI is made wider.

    Letters "G" with font of  7 giving the result as "O" which has a font of 5.

  4) Special Cahracters are not identified for training if the font is less than 2

 5) Letter ' r ' that exists in the same line at different places is not identified the same in both the places"Overspeed     TripThreshold ".

    For the above parameter letter r in the 1st and 3rd instance appears in an individual rectangle but the 2nd Tr appears in single rectangle .

 

Kindly help us in this regard

 

Thanks all in Advance.

0 Kudos
Message 1 of 3
(3,406 Views)

Boy is it difficult to answer questions like this.  You really need to provide more background information.  Where are your images coming from (camera, video signal)?  Do you have a few sample images that will help us understand your problems?  In any case, here are some partial answers:

 

1.  OCR typically requires a minimum font size.  If you are trying to recognize very small characters in a video signal, you will probably need to adjust some of the minimum sizes for OCR to read them properly.

2.  You can set a manual threshold level.  Make sure it is somewhere between the gray and black values and everything should be fine.  If it is the same font, there is no reason to train separately.

3.  Not really clear, but not surprising.  Training several sizes and styles of fonts in the same file can give you bad results.

4.  Same as 1.

5.  You are probably trying to read a font that uses kerning.  This is when one character is squeezed under the edge of the previous character.  Those fonts are very hard to read.  My only suggestion would be training Tr as a single object.  If you train a single character at a time, you can enter multiple characters for a single object.

 

Bruce

Bruce Ammons
Ammons Engineering
Message 2 of 3
(3,396 Views)

You can have a look at this. Mark the above post as solution when answered.

Hope this helps

Shreyas Hebbare
Shreyas Technologies
India
0 Kudos
Message 3 of 3
(3,385 Views)