Signal change detection method used for mine-like objects segmentation in sonar images

In this paper a Signal Change Detection (SCD) method is used and a novel segmentation (SCDS) approach is presented. SCDS is used for segmenting underwater Mine-Like Objects in side-scan sonar images. Generally, SCD is a statistical method used for detecting the time instances where the amplitude levels of amplitude discrete signals change significantly. SCD approximates 1D real signals by choosing a predefined number of jumps of an amplitude discrete signal that achieves maximum similarity, i.e. fitting probability. Changes may occur at unknown time instances with unknown amplitude levels. In this paper, the row and/or column vectors of 2D Sonar intensity images are considered as 1D signals. Two unknown time instances are detected with three amplitude levels, which are calculated as mean values within the detected time indices. Experimental results show that the SCDS approach achieves an efficient segmentation of sonar images in the desired background, shadow (low intensity) and object (high intensity) region.