Using context to detect underwater objects

In the absence of sufficient information about the identity of an object to detect it among other objects there is often extra information in the relationship between objects that can be exploited. This contextual information can help us to characterise the natural set of surrounding objects or even separate objects into different classes directly. Context is widely used to understand and characterise scenery. This paper introduces the use of context to detect underwater objects. Context has not previously been studied for complex sonar imagery and our aim is to design and test measures of context to demonstrate how it can help us to detect and/or identify "unauthorised" objects that are not part of the natural set of objects present in the imagery. This is motivated by some of the challenges that are faced in cluttered sonar environments, or when the objects to be detected are hidden or disguised, for which extra information can be decisive.