Image recognition

Started by
6 comments, last by Replicon 18 years, 10 months ago
Hi there :) This summer I will have to implement real-time image-recognition. My application should be able to recognize simple patterns on flat surfaces, i.e. a pattern made of black and white lego-blocks. I was hoping that some of you have previous experience with this topic, since I'm a newbie in that area :) Any links to scientific papers etc. are most appriciated! I wish you a great summer, Best regards, Roquqkie
Advertisement
Perhaps it helps your search to know that this is in the field of Computer Vision. Use Google and Citeseer (use Google to find Citeseer). Also, I can recommend the book Computer Vision by Shapiro and Stockman; it discusses many image recognition algorithms.

Greetz,

Illco
Consider taking a look at Intel's OpenCV library.
Hi,

My thesis work is about real-time image-recognition, or more specifically pedestrian detection.

First off, there are two books I can recommend.
- Fundamentals of digital image processing, by Anil K. Jain
- Image Processing, Analysis, and Machine Vision, by Milan Sonka, Vaclav Hlavac, Roger Boyle.

Articles are more problem specific, so I'm afraid I don't have any hot tips there for you. A search in a database, such as IEEE or INSPEC might give some search hits.

I would take a look at neural networks too. Might find something there that fits your problem.

Usually there are many solutions to a problem. The hard part is to find the one that works best... :)

Good luck with you research!
Thank you very much, all the three of you! :)

I will look the books up asap.

Skanatic: Which university do you study at?

Best regards,
Roquqkie
I study at Chalmers University of Technology in Gothenburg.
we did basic image recognition in a neuronal networks course. There was a java demo on the net where you could draw paterns that are to be lerned and the query by example. But I don't have the link any more...

There is a up and a down side to this aproch. The recognition is error tollerand, meaning that if you don't have an exact match it would still be recognize. On the other side if the patterns aren deverse enaugh (consider 1, l and I) you will get bad results. -, +, |, x would be no problem.

An other drawback is that the network has to be trained. So you have to know what patterns you want to recognize ahead of time, wich isn't good for all applications. Neuronal networks are fairly easy to implement so it might be worth a shot.

nefthy (who currently can't login. damned password managers, they are almost as bad as cellphones with integrated phonebooks..)
Yeah, I've done some basic image recognition (text recognition actually) using neural nets. Is that how it's generally done, or are there more interesting techniques? I suppose I'll go look inside the book on amazon :-)

This topic is closed to new replies.

Advertisement