Automatic target detection of sonar images using multi-modal threshold and connected component theory

摘要

The aim of this paper is to present a complete progressive development of object detection from underwater acoustic images. Object detection with respect to automatic target detection in underwater autonomous vehicle system is still in a severe problem in context of surveillance and other defense activity. The present work is based on robust method in perspective of segmentation and feature extraction. Underwater acoustic images suffer from typical noise associations and are often of low contrast. In this perspective, a multi-modal thresholding is adopted for automatic segmentation of the images thus obtained and a graph theoretic approach based on connected components is formulated in order to interpret features embedded within the image context. An imaging SONAR is used for carrying out necessary experimental work. The proposed algorithm is executed in comparison with multi-level thresholding and K-means clustering. Effectiveness is established in the context of both running time and quality of processed image as well. The latter aspect is determined by a Figure of Merit (FOM) parameter.