Segment-based bu

In this paper we propose irregular g detection combining optical and interferome features. A Conditional Random Field (CRF) set up on a regular grid of square image compared to a CRF based on irregular image s by a segmentation. We also investigate the pote irregular graph structure in terms of a m formulation of context within the CRF interac shown that computation time is significan results are slightly improved with a graph segments. I. INTRODUCTION The detection and extraction of an obj scene is a core task of computer vision an image analysis. Context information can detection if local features are insufficient be are small, inhomogeneous, partially occlu scattered across a large scene. This is tru remote sensing images of urban scenes b interest like buildings are not placed arbitrari random clutter, but according to certain or (e.g., governed by needs of urban plan locations of buildings and other urban object recreation areas, are mutually dependent consideration of this context can facilitate bui Various approaches have been proposed order to transfer findings of cognitive psycho and to apply them to automatic object detect Many of them view context in a probabi among others facilitates post-processing. Tor an approach for object detection in terre models the relationship between large scale f context and object features probabilistical explicit identification of context-objects. He propose to combine context derived from im single-site features of a sliding window. Th of image segments and then probabilistically to image windows via spatial relations. completely rely on image segments. They ex of regions generated by a multi-level discriminatively learn context. They set up a where the leaves correspond to the image finest segmentation, the root is the entire im