Multi-layer Label Fusion Models

In this chapter, new multi-layer Bayesian label fusionLabel fusion models are proposed for two different change detectionChange detection problems in remotely sensed images. First, a probabilistic model is proposed for automatic change detectionChange detection from airborne images captured by moving cameras. To ensure robustness, an unsupervised coarse matching is used instead of a precise image registrationImage registration. The challenge of the proposed model is to eliminate the registration errors, noise, and the parallax artifacts caused by the static objects having considerable height (buildings, trees, walls, etc.) from the difference image. The background membership of a given image point is described through two different features, and a novel three-layer Markov Random Field (MRF)Markov Random Field (MRF) model is introduced to ensure connected homogeneous regions in the segmented image. Second, we introduce a Bayesian approach, called the Conditional Mixed Markov model (CXM)Conditional Mixed Markov model (CXM), for extracting regions of relevant changes from registered aerial imageAerial image / aerial photo pairs taken with large time differences possibly under different illumination and seasonal conditions. The CXMConditional Mixed Markov model (CXM) model is derived as a combination of a mixed Markov model and a conditionally independent random field of signals. The new approach fuses global intensity histograms with local block-based correlation and contrast features. A global energy optimization process is developed, which can simultaneously ensure efficient local feature selection and smooth, observation-consistent image segmentationImage segmentation. Experiments are shown using real aerial imageAerial image / aerial photo sets provided by the Lechner Knowledge Center of Budapest.