You are here

Unsupervised segmentation of synthetic aperture radar sea ice imagery using MRF models

TitleUnsupervised segmentation of synthetic aperture radar sea ice imagery using MRF models
Publication TypeConference Paper
Year of Publication2004
AuthorsDeng, H., and D. A. Clausi
Conference Name1st Canadian Conference on Computer and Robot Vision
Date Published05/2004
Conference LocationLondon, Ontario, Canada
Keywordsexpectation-maximization (EM), Gamma distribution, image feature, image segmentation, K-means clustering, Markov random field (MRF), mixture model, sea ice, synthetic aperture radar (SAR), texture, unsupervised segmentation

Due to both environmental and sensor reasons, it is challenging to develop computer-assisted algorithms to segment SAR (synthetic aperture radar) sea ice imagery. In this research, images containing either ice and water or multiple ice classes are segmented. This paper proposes to use the image intensity to discriminate ice from water and to use texture features to separate different ice types. In order to seamlessly combine spatial relationship information in an ice image with various image features, a novel Bayesian segmentation approach is developed Experiments demonstrate that the proposed algorithm is able to segment both types of sea ice images and achieves an improvement over the standard MRF (Markov random field) based method, the finite Gamma mixture model and the K-means clustering method.