Abstract:

A boundary detection algorithm for synthetic aper- ture sonar (SAS) images that draws upon a popular superpixel formation algorithm is detailed and tested against a set of SAS images containing a variety of common seabed categories. Textural clues are gathered using a novel DP clustering algorithm that replaces the traditional K-means operator used in texton clustering. The superpixel formation algorithm uses texture and image pixel intensity cues as the input to a K-way normalized cuts algorithms. Superpixel boundaries computed by the normalized cuts algorithm are then merged using a Jensen-Shannon (JS) divergence metric. These larger regions boundaries resulting from superpixel merging contain homogeneous seabed charac- teristics and are useful for follow on image segmentation and environmental parameter estimation. To illustrate the algorithm’s performance, boundary detection results are plotted in receiver- operator characteristic curve plot and selected oversegmented images are displayed at various JS divergence thresholds.

Links:

Citation:

J. T. Cobb and A. Zare, “Boundary detection and Superpixel formation in synthetic aperture sonar imagery,” in 3rd Int. Conf. SAS and SAR, 2014. 

@InProceedings{cobb2014boundary,
author = {Cobb, James T. and Zare, Alina},
title = {Boundary detection and Superpixel formation in synthetic aperture sonar imagery},
booktitle = {3rd Int. Conf. SAS and SAR},
year = {2014},
month = {Sept.},
}