So I'm trying to setup some stuff with opencv for image segmentation and then classification. I was thinking about using a neural net or an SVM for the classification. I already have something going for a mixture of gaussian system and a meanshift algorithm to do the segmentation. My goal is to eventually classify say my roomate from myself.
However, I'm looking for ideas on how to create a system that will keep learning and/or adapt over time. Would the best way to go about this just keep training the SVM upto a certain size? If so, what is a good size of input feature vectors (assuming rectangular, segmented regions of foreground binary images)?
[link][2 comments]