基於交叉雙邊濾波的立體影像去霧研究

Abstract

本論文旨在針對立體影像並採用交叉雙邊濾波器進行去霧。在此之前，已經出現大量的去霧演算法。然而這些去霧演算法多是針對單張影像。倘若，直接對立體影像分別進行單張影像去霧，容易導致去霧結果不一致的情況，進而影響使用者觀看立體影像的體驗。 在本論文中，除了對立體影像去霧，同時也預估場景深度。本論文的重點在於根據立體影像產生場景深度，並以此場景深度結合交叉雙邊濾波器進行去霧。此方法可以避免去霧結果不一致的情況。最後，本論文的測試結果顯示，相較以往的方法可以得到更好的結果。

In this thesis, we present a novel dehazing approach for stereo images based on cross bilateral filtering. Numerous dehazing algorithms have been proposed before. Nevertheless, most of the dehazing algorithms are proposed for a single image. This will produce inconsistent results if dehazing stereo images iteratively. We simultaneously estimate scene depth and dehaze the stereo images. The proposed approach is based on the observation of depth cues in the stereo images. The main idea of using depth cues is to avoid inconsistent results and of using the cross bilateral filter is to preserve shape details. The results show that the proposed approach can get better results compared with the previous methods.