OPTICAL CHARACTER RECOGNITION within PREMIER AUTOMATION . . .
All --
An IMAGE-PRO PREMIER USER has come to me with a challenging project.
One of the challenges is that a LABEL within an IMAGE contains information they would like included in the DATA FILE generated by an analysis of the image. To satisfy this user, I would like to propose a PREMIER AUTOMATION that can extract a string of CHARACTERS from the LABEL so that the INFORMATION on the LABEL can be included with MEASUREMENTS made within the image.
I have searched here under
OCR
OPTICAL CHARACTER
and within the MEDIA CYBERNETICS SOLUTIONS ZONE.
So far I have not seen anything that applies.
Is this a feature of PREMIER or demonstrated within an EXAMPLE APP or is this outside the FEATURES and FUNCTIONS of PREMIER?
Thanks.
-- Matt
An IMAGE-PRO PREMIER USER has come to me with a challenging project.
One of the challenges is that a LABEL within an IMAGE contains information they would like included in the DATA FILE generated by an analysis of the image. To satisfy this user, I would like to propose a PREMIER AUTOMATION that can extract a string of CHARACTERS from the LABEL so that the INFORMATION on the LABEL can be included with MEASUREMENTS made within the image.
I have searched here under
OCR
OPTICAL CHARACTER
and within the MEDIA CYBERNETICS SOLUTIONS ZONE.
So far I have not seen anything that applies.
Is this a feature of PREMIER or demonstrated within an EXAMPLE APP or is this outside the FEATURES and FUNCTIONS of PREMIER?
Thanks.
-- Matt
0
Comments
This means that Premier has pattern matching functionality. Using pattern matching agains a set of reference images (alphabet) you can find what the letter is.
Pattern matching controls are on the Extras tab, you have to activate this in the application options. Check the help for Pattern Matching for more details.
Yuri
Thank you for your suggestion.
After turning on the EXTRAS RIBBION OPTION, I worked with this PREMIER FEATURE.
The EXAMPLE IMAGE taken from the INTERNET has 3 Rs within it.
The best TRAINING / SETTINGS that I got found two of the Rs.
I have included a SCREEN CAPTURE that found TWO Rs and T3 FALSE POSITIVES.
Can you suggest a better way to do this?
Thanks.
You have to know limitation of cross-correlation based Pattern Matching, it will match only objects of the same shape, it will not work on letters that have different font or style (Bold, italic).
If you want to handle orientation, scale, translation invariant recognition, you can take a look at Fourier Descriptor. There is a post on the forum that demonstrates how it can be used for shape classification:
http://forums.mediacy.com/discussion/103/how-to-use-fourier-descriptor-for-object-shape-classification/p1
Yuri
I have an idea, not so simple, but I think, you can do this in the Premier script. Steps:
1) Extract portion of the image with the label in the separate file,
2) Recognize characters using http://www.verydoc.com/raster-to-text-ocr/image-to-text-ocr-converter.html from the command line (extra cost, but worth to look for free analogs).
3) Read text from the result file.
Thanks,
Nikita.
Thank you for your responses on 19-FEB and 22-FEB
I've been on vacation for a week but I'm back at my desk and working to get my projects back in motion.
I'll follow up on your recommendations ASAP and reply here with the results.
Thanks again.
-- Matt