Hierarchical image-region labeling via structured learning

Abstract

We present a graphical model which encodes a series of hierarchical constraints for classifying image regions at multiple scales. We show that inference in this model can be performed efficiently and exactly, rendering it amenable to structured learning. Rather than using feature vectors derived from images themselves, our model is parametrised using the outputs of a series of first-order classifiers. Thus our model learns which classifiers are useful at different scales, and also the relationships between classifiers at different scales. We present promising results on the VOC2007 and VOC2008 datasets. © 2009. The copyright of this document resides with its authors.