Estimação automática de parâmetros de modelos para restauração de imagens de cenas subaquáticas

Abstract

A large number of techniques to extract information contained in images have been developed through the years. However, these techniques are based on the premise that objects and observer are immersed in a transparent medium, as the air for example. It is assumed that light rays travel through the mediumwithout any alteration. However, there are scenarios where the medium may change the intensity and the direction of the light rays, as underwater environments.This dissertation presents a study about viewing of underwater environments with natural light problem. The degradation in underwater images visibility is due to mainly by \emph{Attenuation} and \emph{Scattering}. A survey of the last approach to handle the underwater images problem and a novel methodology are presented. The problems observed in underwater images are related to by the interaction of light with water. We present a methodology to enhance the visibility of underwater scenes automatically, combining a physical model and Comptuer Vision algorithms. The approach uses a pair of cameras to acquire images, a stereo system to estimate the three-dimensional structure of the scene and two techniques to estimate the parameters of the model for light propagation in underwater scenes.Simulations and experiments were performed to evaluate the limitations and the qualities of restoring underwater images using the methodology presented. Using qualitative and quantitative metrics we obtained promising results.