Implicit scene context for object segmentation and classification

Abstract

In this paper, we propose a generic integration of context-knowledge within the unary potentials of Conditional Random Fields (CRF) for object segmentation and classification. Our aim is to learn object-context from the background class of partially labeled images which we call implicit scene context (ISC). A CRF is set up on image super-pixels that are clustered into multiple classes. We then derive context histograms capturing neighborhood relations and integrate them as features into the CRF. Classification experiments with simulated data, eTRIMS building facades, Graz-02 cars, and samples downloaded from Google™ show significant performance improvements. © 2011 Springer-Verlag.