Structured prediction for urban scene semantic segmentation with geographic context

M. Volpi, V. Ferrari

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract / Description of output

In this work we address the problem of semantic segmentation of urban remote sensing images into land cover maps. We propose to tackle this task by learning the geographic context of classes and use it to favor or discourage certain spatial configuration of label assignments. For this reason, we learn from training data two spatial priors enforcing different key aspects of the geographical space: local co-occurrence and relative location of land cover classes. We propose to embed these geographic context potentials into a pairwise conditional random field (CRF) which models them jointly with unary potentials from a random forest (RF) classifier. We train the RF on a large set of descriptors which allow to properly account for the class appearance variations induced by the high spatial resolution. We evaluate our approach by an exhaustive experimental comparisons on a set of 20 QuickBird pansharpened multi-spectral images.
Original languageEnglish
Title of host publicationUrban Remote Sensing Event (JURSE), 2015 Joint
PublisherInstitute of Electrical and Electronics Engineers (IEEE)
Number of pages4
ISBN (Print)978-1-4799-6652-3
Publication statusPublished - 1 Mar 2015

Keywords / Materials (for Non-textual outputs)

  • geophysical image processing
  • geophysical techniques
  • image segmentation
  • land cover
  • remote sensing
  • QuickBird pansharpened multispectral images
  • geographic context
  • land cover classes
  • land cover maps
  • pairwise conditional random field
  • random forest classifier
  • structured prediction
  • urban remote sensing images
  • urban scene semantic segmentation
  • Accuracy
  • Context
  • Context modeling
  • Image segmentation
  • Semantics
  • Training
  • Vegetation


Dive into the research topics of 'Structured prediction for urban scene semantic segmentation with geographic context'. Together they form a unique fingerprint.

Cite this