Uma abordagem automática para restauração de imagens de cenas subaquáticas

• Ano de defesa: 2010
• Autor(a) principal: Wagner Ferreira de Barros
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: visão computacional; visão sub aquática; Restauração de imagens; Visão por computador; Computação; Processamento de imagem Técnicas digitais Conservação e restauração
• Link de acesso: http://hdl.handle.net/1843/SLSS-85YFWZ

This work describes a fully automatic methodology that restores images acquiredfrom underwater environments and makes it possible to apply classical computervision and image processing algorithms to underwater images.The restoration process requires a single image as input and returns an imagewhere the eects of the participating media are significantly reduced.The proposed methodology aims at joining the benefits from restoration methodsthat operate both in the scene and the image domains, and does not require the useof additional equipment for image acquisition besides single camera.The image formation model is based on Jae-McGlamerys model including themain degradation eects due to light attenuation and scattering in the medium.The restoration process uses a non-linear optimization method that adjusts themodel's parameters while maximizing a function that adequately captures a set ofimage features describing the image quality. We have selected four quality indices:global contrast, integrity edge, the blurring near the edges and an estimation of theadditive noise. A preprocessing procedure has been developed, wich makes therestoration process more faithful to the physical model used in the image formationmodel. The image preprocessing procedure is split into two parts that provide twoimportant parameters used in our restoration algorithm: a saturation map and aweighting vector to perform chromatic compensation.The method was experimentally validated with images acquired from simulationsof underwater scenes, images acquired in laboratory and outdoor underwaterenvironments. The methods performance is evaluated by comparing the applicationof a real visual task to raw images, images restored by our methodology andby two image restoration algorithms. The results shown that the proposed methodologyproduces images with better visual quality and feature description than theother evaluated methods, in all experiments.