The underwater world is undoubtedly a difficult and challenging environment and investigating it requires huge efforts from both the technological and the data-processing point of view. The inherently structureless environment and the difficulties caused by the nature of the propagating medium have limited the types of sensors to be used for data acquisition; thus they have also restricted knowledge of the submerged world.
However, interest in exploring and, above all, "understanding" underwater scenes is growing and may include many applications in the physical, biological, geological, archaeological, and industrial fields.
In particular, this interest is currently aroused by the introduction of new and smarter imaging sensors that allow one to recognize and interpret visualized scenes directly, rather than having to infer structures from raw signals. Besides, advanced image-processing methods, devised in the computer vision and pattern recognition areas, have further contributed to improving the scenario.
Typically, both acoustic and optical sensorial systems are used to sense the underwater environment. More specifically, several types of sonar systems (e.g., side-scan sonars, forward-looking sonars, acoustic cameras, and multibeam echo-sounders) and videocameras (possibly with increasing capabilities, like high sensitivity to low levels of luminosity) are exploited. Laser-based systems are also utilized, though less frequently than the abovementioned ones.
Apart from the difficulties inherent in the sensorial system used, several typical problems are usually associated with the underwater environment. Therefore, before any high-level processing is performed, underwater data are generally preprocessed to reduce the amount of noise and also to correct the attenuation and the geometric distortion that may affect the images acquired.
Subsequently, further problems are posed by image segmentation, image classification, and recognition of the seafloor or, generically, of observed objects. Such operations can be handled by using both acoustic and optical sensors. Three-dimensional reconstruction of surfaces or objects and registration of maps are other important processes useful in many applications concerning underwater vehicle navigation.
At present, the above problems are faced by exploiting theories and techniques provided by the pattern recognition and computer vision areas, rather than by designing an ad hoc method for each specific problem.