A benchmarking study on single image dehazing techniques for underwater autonomous vehicles

Abstract

Enhancing the underwater images is of utmost importance for autonomous underwater vehicles. This kind of robots usually have to deal with highly degraded images from which it is extremely difficult to accurately find, recognise or manipulate targets of interest. For this reason, a single image dehazing fast enough to run in a real time system would be an important tool facilitating image processing. In this paper, an study of different dehazing techniques is presented, experimenting with the most suitable algorithms for this context. A benchmark is described testing dark channel prior based methodologies establishing an objective evaluation of these techniques from a real time application perspective.