Epitomic representation of natural signals
Add to your list(s)
Download to your calendar using vCal
If you have a question about this talk, please contact Phil Cowans.
Many natural signals such as images, audio and genetic sequences have repeating fragments that can be compactly expressed using, what we call as, an epitome. An epitome of a signal is its compact representation that is learned using overlapping segments or patches of various sizes from the signal. In this talk, I will explain a generative model for images using epitomes. When the epitomic representation is used in a hierarchical generative model of image formation, appropriate inference algorithms can be derived to extract epitomes from a single image or a collection of images and simultaneously perform various tasks such as image segmentation, motion estimation, super-resolution and image denoising.
Time permitting, I will preview recent work by colleagues on extending this to videos.
Joint work with Nebojsa Jojic (Microsoft Research) and Brendan Frey (University of Toronto)
This talk is part of the Inference Group series.
This talk is included in these lists:
Note that ex-directory lists are not shown.
|