Image Recognition Library

Question

I wonder whether there is any competent image recognition library ( preferably open source) in .Net? I have an application that scans pages and documents in batches. In order to separate one document from another ( for these documents could have multiple pages), I plan to ask users to put a page, with a say, polar bear image on it to separate between the documents. So my applications will take all the scanning, checks whether the page contain the polar bear, if yes, then call it the end of the first document.

I know about OCR, where the application will recognize the text, but what about the equivalent in images?

Answer 1

I don't have an answer to the general question about image recognition, but given your specific requirement, might it not be easier to substitute a simpler recognition domain?

Specifically, on the page that you ask users to put between documents, you could include a nice chunky bar code as well as the polar bear (or whatever) image. Then you could check for the barcode -- which is easily recognisable -- rather than having to recognise the polar bear image per se.

Answer 2

I am not sure I exactly understand what you are trying to do, but you might check out these questions:

• an-easy-way-tool-to-compare-images-pixel-for-pixel-in-different-formats
• removing-duplicate-images
• are-there-any-ok-image-recognition-libraries-for-net
• comparison-of-bmp-files

If you are looking to implement something yourself to compare an image to a segment of another image. You can do that by using template matching. You can use OpenCV which also has a .NET wrapper

Answer 3

Perhaps worth looking at QR Codes?

Answer 4

Image recognition is hard mainly because every different application requires a different solution. For example, trying to identify a single black symbol on a white background requires an entirely different technique than trying to identify all the people in a colour photograph where some people are overlapping and there is scenery in the background. Some factors that would influence what method to use would be:

• Type of image: symbols/characters, everyday photographs, medical images, shapes, etc.
• Colour: Black and white, grayscale, colour
• Object of interest: faces, people, shapes, symbols, general objects, fingerprints, irises, tumors, etc.
• Training: Do you have a training set of data?

So really what you need to do is tell us more information about the types of images you are trying to recognize so that we can tell you what libraries and what methods would be best to try.

Answer 5

Maybe this framework could be of help:

AForge.NET Framework