Hi all, I'm familiar with some of the state of the art methods for detecting classes of objects in an image (ie, Cars, humans, CIFAR-classes, etc...).

However, if I am going to try to recognize a specific object in an image instead of a class of objects, it seems like a much simpler problem. Suppose I have an object in my room and I want my system to recognize this object when it is in an image (preferably with some invariance to rotation/lighting conditions), are there any algorithms that can take relatively few input examples of my object and then learn to recognize it in images? Unless I'm remembering wrong, I thought I had seen algorithms that could do this before...

Any help pointing me in the right direction would be appreciated. Thanks!

asked Feb 25 '13 at 13:20

Adam%20Fischer's gravatar image

Adam Fischer
16112

I'm not a vision person, but my impression is that any method that works for classes of objects should work for the same object if trained on that kind of data and tuned well.

(Feb 26 '13 at 13:44) Alexandre Passos ♦

I don't think the problem pointed out here pertains to simple recognition of an object in an image. It is a lot more complex as it also involves focus of attention to the object of interest.

To my knowledge, the present day deep learning methods (which are the state of the art models) are still focusing on feed-forward models. With in this scene segmentation is the closest that I could think of that is related to the problem here. Check http://yann.lecun.com/exdb/publis/pdf/farabet-pami-13.pdf

Another more recent paper related to attention is http://www.eng.uwaterloo.ca/~jbergstr/files/nips_dl_2012/Paper%2038.pdf

(Feb 26 '13 at 14:52) Rakesh Chalasani

One Answer:

I think this is a standard image processing problem rather than machine learning

see the SIFT paper http://www.cs.ubc.ca/~lowe/papers/ijcv04.pdf... figures 12 and 13 for example

answered Feb 28 '13 at 18:45

SeanV's gravatar image

SeanV
33629

Your answer
toggle preview

powered by OSQA

User submitted content is under Creative Commons: Attribution - Share Alike; Other things copyright (C) 2010, MetaOptimize LLC.