Detection of small objects in sidescan sonar images based on POHMT and Tsallis entropy

Abstract

SideScan Sonar (SSS) systems are able to provide near-photographic high-resolution images of underwater areas. However, automatic object detection, especially for small objects detection in such images is still difficult because of the inherent speckle noise as well as the various shapes of small objects. In this paper, by using both the Percentage Occupancy Hit-or-Miss Transform (POHMT) and the Tsallis entropy, a novel method for detecting small objects in SSS images is proposed. On one hand, the proposed method adopts the morphological POHMT to enhance faint small objects. On the other hand, the Tsallis entropy which considers the long-range interactions among particles in a system is employed to determine the optimal threshold of the POHMT. The proposed method has been evaluated on two groups of SSS images. For the first group, the average detection rate and false alarm rate of the proposed method are about 86.42% and 10.58%, respectively. For the second group, the average detection rate is 100% while the average false alarm rate is 6.25%. These results show that the proposed method performs better than the existing techniques such as the traditional hit-or-miss transform, the multilevel Tsallis entropy, and the accumulated cell average-constant false alarm rate.