UVW 2013 - Underwater Vision Workshop

There are many challenges associated with processing images captured underwater. Natural scene illumination may be very poor, and there is often little regular structure with which to delineate objects. Additional challenges are introduced by strong wavelength dependent attenuation that limits the effective range of optical imaging in realistic settings to a few meters. This attenuation is the dominant cause of the colour imbalance often visible in underwater images. In practice, there are many cases in which imaging areas of interest underwater requires collecting thousands of images for adequate resolution and quality. In shallow waters, the refraction of sunlight on surface waves and ripples can be problematic, while in deep waters the imaging system needs to carry its own moving light sources resulting in changing illumination in the scene. State of the art camera calibration methods are complex and most practitioners use methods for camera calibration and distortion compensation that do not fully account for refraction of light through the air-viewport-water interface. All of these effects present unique difficulties when working with underwater imagery. Approaches that integrate or couple geometric, radiometric and semantic understanding are likely to be necessary for robust algorithms that will enable a broad range of real-world applications.
Despite these challenges, there have been many recent advances in the processing of imagery from underwater scenes. These advances have implications in a diverse range of application areas, including marine ecology, archaeology, geology as well as industrial and defence applications. This workshop will focus on aspects of underwater-related vision processing, including but not restricted to:
physics based vision, hyperspectral imaging, and plenoptics,
underwater camera calibration, refractive concerns and lighting/camera configuration,
color correction and illumination compensation,
multi-view geometry,
mapping, localization and SLAM,
structure-from-motion and visual odometry,
acoustic imaging and processing techniques,
segmentation,
image classification, object discovery and recognition, scene understanding, and semi-supervised and unsupervised learning for sparsely-labelled or unlabelled data,
methods that are robust to heavy-tailed class distributions, and label/annotation inconsistencies.