Graph Cut based Inference with Co-occurrence Statistics

DMCA

Graph Cut based Inference with Co-occurrence Statistics

Cached

Download Links

by Lubor Ladicky , Chris Russell , Pushmeet Kohli , Philip H. S. Torr , Oxford Brookes

Citations:

100 - 13 self

BibTeX

@MISC{Ladicky_graphcut,
    author = {Lubor Ladicky and Chris Russell and Pushmeet Kohli and Philip H. S. Torr and Oxford Brookes},
    title = {Graph Cut based Inference with Co-occurrence Statistics},
    year = {}
}

OpenURL

Abstract

Abstract. Markov and Conditional random fields (CRFs) used in computer vision typically model only local interactions between variables, as this is computationally tractable. In this paper we consider a class of global potentials defined over all variables in the CRF. We show how they can be readily optimised using standard graph cut algorithms at little extra expense compared to a standard pairwise field. This result can be directly used for the problem of class based image segmentation which has seen increasing recent interest within computer vision. Here the aim is to assign a label to each pixel of a given image from a set of possible object classes. Typically these methods use random fields to model local interactions between pixels or super-pixels. One of the cues that helps recognition is global object co-occurrence statistics, a measure of which classes (such as chair or motorbike) are likely to occur in the same image together. There have been several approaches proposed to exploit this property, but all of them suffer from different limitations and typically carry a high computational cost, preventing their application on large images. We find that the new model we propose produces an improvement in the labelling compared to just using a pairwise model. 1

Keyphrases

co-occurrence statistic graph cut computer vision local interaction global object co-occurrence statistic standard pairwise field conditional random field high computational cost global potential several approach different limitation standard graph cut algorithm random field recent interest little extra expense possible object class new model pairwise model large image image segmentation

DMCA