Multiscale Random Field Model for Bayesian Image Segmentation

摘要

The Army's Land Condition Trend Analysis (LCTA) program collects both space-basedremotely-sensed data and ground-level data for natural resource inventory and evaluation. Coupling remotely sensed digital data with traditional ecological ground data could help Army land managers inventory and monitor natural resources. This study used LCTA data sets to test image segmentation algorithms that may be used to interpret remotely sensed digital data. Many approaches to Bayesian image segmentation have used maximum a posteriori (MAP) estimation in conjunction with Markov random fields (MRF). This study developed a new approach to Bayesian image segmentation that replaces the MRF model with a novel multiscale random field (MSRF), and replaces the MAP estimator with a sequential MAP (SMAP) estimator derived from a novel estimation criteria. Together, the proposed estimator and model result in a noniterative segmentation algorithm that can be computed in time proportional to MN, where M is the number of classes and N is the number of pixels. Results show that the SMAP algorithm improves classification accuracy when applied to the segmentation of multispectral remotely sensed images with ground-truthed data, especially with the introduction of mixed signatures to represent the spectral information about the classes. Land Condition Trend Analysis (LCTA), Multiscale Random Field (MSRF), Bayesian image segmentation, GRASS.