Image Processing in Side Scan Sonar Images for Object Detection and Classification

This paper presents recent and planned activities in the area of computer aided detection and classification (CAD / CAC) of mine like objects (MLOs) at the FWG with assistance of FU-Berlin and FGAN-FOM. These investigations are intended to support software for the analysis of side scan sonar images by an operator and to contribute to automatic target recognition (ATR) software in case of autonomous UUVs. The object detection and classification process is divided into a couple of processing steps starting with pre-processing and screening for regions of interest over the reduction of false positives and object classification up to the fusion of detection and classification results of different algorithms. A first CAD system was presented 2008 [2] including four different screening methods (statistical analysis, highlight / shadow analysis, correlation with a 1dtemplate and a modified Maximally Stable Extremal Regions (MSER) approach) as well as four different algorithms for the reduction of false positives (snake algorithms, correlation with a set of 2d-tamplates and iterative segmentation). In the meantime investigations have been made to study the influence of pre-processing on the detection and classification process, additional algorithms have been implemented (e.g. k-means and higher order statistical based segmentation, co-occurrence matrixes based entropy and energy, ...) and first results were achieved. All algorithms have been or will be tested using a SSS data currently covering 25 km of the seafloor. These data were in part collected by the SeaOtter MK I AUV from Atlas Elektronik gathered in the Baltic Sea and the Mediterranean Sea.