On Sonar Image Processing Techniques for Anomaly Detection in Underwater Constructions

Abstract:
To detect small dimensions in sonar images of underwater constructions reliably advanced image enhancement and reconstruction techniques are required. Denoising techniques are mandatory before subsequent procedures (image segmentation, feature extraction, and target classification) are applied. We present diffusion model based approaches employing both Perona-Malik and variation models with different regularizations. Thresholding and EM based segmentation algorithms are applied for the objects identification in sonar scenes. The performance of thresholding procedure is severely affected by the threshold selected. For evaluating the performance, experiments on synthetic aperture sonar (SAS) images are conducted.