Best way to integrate mutiple classifiers into a single system

asked 2014-09-17 11:34:54 -0500

sip gravatar image

I am a bit of a noob so please excuse the question if it is not phrased correctly.

I want to build a simple detector for some objects and want to be able to run all of these against a new image to see if the image contains any of the items.

currently as i understand it I would need to do something like orb or sift feature detection on the item and build a classifier from this.

the next step is where i am struggling. if i have for example 1000 items for which i have descriptors how would i combine these in order to be able to detect any of the items within a new image?

I have seen convolutional neural nets mentioned aswell as deep learning and vocabulary trees but im unsure what direction to take in terms of learning next

In addition to the above problem I also in some cases have more than one view of an item. for example a front on view and then a side view. or even multiple angles. Is it possible to combine a full classifier that would detect an object from multiple angles into a larger system for detecting lots of items?

sorry foer the long description and thanks in advance for the help

edit retag flag offensive close merge delete

Comments

If you have multiple descriptors for items, perhaps you run each of those descriptors on your image, possibly within a ROI created by a sliding window?

To deal with the angles thing, perhaps you can just run each of the classifiers you have for each angle, and have each classifier say yes or no, and if X classifiers say yes then you can say you have that object?

These are just suggestions, so I figured I'd put them in a comment rather than an answer. I don't know anything about convolutional NN or voc. trees sorry.

ejehardenberg gravatar imageejehardenberg ( 2014-09-17 11:40:28 -0500 )edit

Thanks for the feedback. yes i agree the classifiers for different angles will essentially be just standalone classifiers and the fact they depict the same object is not important. the resulting detection would give the same answer but it would not be linked in any other way.

sip gravatar imagesip ( 2014-09-18 12:28:50 -0500 )edit