ASM Community

Does anyone know about a free/commercial library which can be used for image recognition / pattern matching?

Thing is, that I need to code a software which takes a digicam picture and recognizes which part (bumper parts... I work in the automotive industry) the picture shows.
The position of the cam and the part is fixed because the picture is made while the part is in a machine. The lighting on the part can be differ a bit.

I am currently trying to make something like this myself, but am curious if I could save some work :)

My current approach is to convert the image to black/white (monochrome) and look for about 40 fixed points (pixels) where the picture must be black/white. Would you know a better way of doing it?

Well, I dont need a motion detection, I need something like a "object" detection.
I already hacked something out today and I might get a first working thing over the weekend :/

I know that I'm on a thin line with what I'm about to post, but anyways.... :rolleyes:

On Hiro's forum, that we're not supposed to mention, there was a thread awhile back about this. Someone was wondering, out of curiosity not evil intentions, if it's possible to programmatically compare the letter/number combinations that you have to verify before signing up for free webspace, email, etc. Those grainy little pics that look like the color blind tests on those types of sites. Anyways, you may want to look for that post and see if anyone came up with anything.

sounds very hard :|

I think I have an old dr. dobbs on noise reduction on BW images lying around here, I'll see if I can find it for you.

to get a sharper image :p

bazik, you plainly posted what you're trying to do. I realize that you don't care about letters and numbers. It just brought up the memory of that project on the other board. Both of you are trying to identify parts of images, and I didn't think it would hurt to check out their progress as they might have come up with something that you can use. Those guys can be pretty ingenious at times. To each his own I guess. -shrugs-

There is some rudementary tutorials on Generations5, regarding object detection (Very hard, and undeveloped field ~ well for non-propriatary work). I have a need for the same but for entirely different purposes. Perhaps we could colaborate and make a summer project out of it ~ (short term, ie. 3 weeks, is way too busy for me).

Your lucky in the sense of the camera angle.I assume the 'bumpers' are more or less in the same basic outline from test photo to test photo. My needs are entirely random, with random # of shapes to outline too.

I have an app called QuickStitch by Enroute Imaging that came with my digital camera that allows me to stitch digital pics that have 30-50% overlap together to make a panaramic image. It uses pattern recognition to line up the overlapping areas and does an amazing (to me) job of it even when the images are not aligned very well. Perhaps you can find something useful by searching for "image stitching", "panaromic images" or some such terms...

Use edge detection. Subtract an image shifted diagonally down-right by one pixel from original image (both monochrome), and you have edges. Convert edges to lines, simplify positions (perhaps just shrink image to rather small dimensions), then check for coordinates of lines to preset ones.

Sorry to bump this old thread back to the top, but I just found that Intel has an open-source image / outline recognition library. Thought it'd be of interest :)

http://www.intel.com/research/mrl/research/opencv/overview.htm