Achieving Turbidity Robustness on Underwater Images Local Feature Detection

Methods to detect local features have been made to be invariant to many transformations. So far, the vast majority of feature detectors consider robustness just to over-land effects. However, when capturing pictures in underwater environments, there are media specific properties that can degrade the visual quality the captured images. Little work has been made in order to study the robustness that the popular feature detectors have to underwater environment image conditions. We develop a new dataset, called TURBID, where we produced real seabed images with different amounts of degradation. On this dataset, we search over multiple feature detectors from the literature to indicate the ones with more robust properties. We concluded that scale-invariant detectors are more robust to degradation of underwater images. Finally, we elected Center Surround Extremas, KAZE, Difference of Gaussians and the Hessian-Laplace as the best detectors for this environment on all tested scenes.