Reconstruction and segmentation of underwater acoustic images combining confidence information in MRF models

This paper describes a technique for the integration of con"dence information using Markov random "elds'models to improve the segmentation and reconstruction of three-dimensional acoustical images. A range image in which each point is associated with a reliability measure, namely, its `con"dencea, is directly provided by acoustic image formation process. In this paper, range and con"dence images are modelled as Markov random "elds over which several energy formulations are devised to exploit both types of data, leading to an accurate reconstruction and segmentation of such images. Results show the better performances of the proposed approach as compared with classical methods disregarding reliability information. ( 2001 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.