A novel automatic segmentation algorithm for sonar imagery

Segmentation of underwater objects using sonar imagery is complicated by the variability of objects, noises, and background. Through analyzing the features of sonar imagery, we propose a new algorithm which can carry on automatic segmentation on sonar imagery. After noise reduction and image normalization, we adopt the self-adaptive variance algorithm and the fractal dimension algorithm to segment the high-light areas and the shadow areas respectively and thus complete initial segmentation. Then according to the initial segmentation results, we carry on estimation of the initial parameters of the MRF (Markov random field) models and the following we conduct ICE (iterative conditional estimation) algorithm based on the MRF theory to obtain the final precise segmentation results. In the last part, experiments are conducted to demonstrate the feasibility and effectiveness by the data detected practically. This segmentation algorithm is based on analyzing the structure of objects in sonar imagery and works well in the sonar imagery.