09-18-2011 08:50 PM
Good day NI Gurus,
I would like to inquire about an OCR technique I could use regarding my application. Basically, The characters don't have definite spacing. Some are apart from other character while some are attached to others. Because of this, I was forced to assign an OCR ROI in per character basis as specified in the attached images. I've had success in reading all the characters in an IC but when another IC is about to be read, the result is disastrous. Hope you could give me an advise regarding this.
Thank you very much,
Andrei K.
09-19-2011 04:54 PM
Hello Andrei
Thank you very much for getting in touch with us. I would be more than happy to assist you with this issue. Just so I am totally clear, can you describe what I should be looking for in each individual image. From what I gather, it looks like when the characters overlap, they aren't distinguished. Is there a reasonabley small number of characters that will be printed close to eachother like that? If so, you may want to define some more characters. What is going on in the bigger picture? Why are the letters being printed like this? Is there any solution to spacing the letters out more? Please let me know!
09-19-2011 07:45 PM
Hello Mr. Greg,
Thank you very much for your kind response. The large image shows "2" IC's (lower left) of same kind in full view. I am also reading data from other IC's but this one is really difficult to read. Please disregard image "IC5clear". What is required from us is to read the texts from these 2 IC's. For the moment anyway, we can focus on "2" IC samples "small1" and "small2" and off course we should expect for more other IC's. I am not familiar with the IC or how the characters are printed but what we need to do is to read every character in the IC regardless of how are they positioned or spaced for that matter. As for my observation in terms of size, an "88" character would be equal to "111". I do hope we could space out the characters in the IC but that would not be possible. Hope we could find solution to this problem. Again, Thank you very much.
Sincerely,
Andrei K.
09-20-2011 02:57 PM
Hello Andrei
Thank you very much for your response! I would recommend experimenting with Threshold/Size/Options values in Read/Verify Text Setup. For images where numbers are stamped on top of eachother, this may be more difficult to accomplish in VBAI alone and may require some LabVIEW intervention. When the image contains characters that are stamped on top of eachother or so nearby that their boundaries connect, the software views this as one character. Is there a small set of characters that are placed side by side because if so, you may just want to create some more templates. This process is difficult as some of the characters are even difficult to discern with the human eye when the they are stamped on top of eachother. Please let me know how you would like to proceed.
09-20-2011 07:23 PM
Hello Mr. Greg,
Thank you very much. I've already experimented with every setting possible I can utilizing Vision Assistant and Read/Verify Text Setup. Please refer to my sample program.
Sincerely,
Andrei K.
09-20-2011 07:25 PM
Attached is the sample image "IC5clear2". I'll send more samples later.
Thanks,
Andrei K.
09-20-2011 09:46 PM
Hello Andrei
Just out of curiosity, what percentage of the images end up failing in your inspection? Is behaviour something that occurs in the majority of images collected?
09-21-2011 12:57 PM
Hello Andrei
Thank you very much for the sample inspection and images. Can you please let me know how I can best assist you with this issue? Is there something specific about the examples or inspection you want me too look at? I am happy to help so please let me know so I can best assist you.
09-21-2011 07:45 PM
Hello Mr. Greg,
What we need is to teach and read all the specified characters in the IC. The ROI and reading should be stable for all IC's. Thank you very much.
Sincerely,
Andrei K.
09-22-2011 09:28 AM
Hello Andrei
Can you provide a table of ranges of values that you have tried for the different parameters? That would be really helpful as it would allow me to focus on areas that may not have yet been explored. Thank you very much!