Detection of objects and their shadows from acoustic images of the sea floor

We propose a novel method for the automatic detection of salient regions in acoustic images. Two types of regions are of interest, namely objects and their cast shadows. Unlike shadows cast by objects in optical images where the light source model is often unknown or uncontrolled, acoustic shadows provide useful information about the object´s shape and are therefore considered as representing salient regions. Our method is model-free, as it makes no assumption about the nature of the objects (and their cast shadows) present in the image. Both detection algorithms use the watershed transform as an initial step. Shadow detection is based on texture analysis, while objects are detected by using a modified watershed transform, namely the marker-based watershed transform.