白天與夜晚影像霧霾及煙霧消除演算法之研究

Haze Removal in Daytime and Nighttime Scene and Simple Image Desmoking by Haze Image Model

王彙智(Hui-Chih Wang)

指導教授 ： 貝蘇章

國立臺灣大學/電機資訊學院/電信工程學研究所/碩士(2017年)

https://doi.org/10.6342/NTU201701442

摘要

本篇論文主要探討除霧演算法在各方面之應用，包括在白天與夜間環境光源下除霧，進一步修改除霧演算法使其適用於被煙霧遮蔽之影像。論文第一個部分主要探討的是白天環境光源下的除霧演算法，基於現有的除霧演算法，dark channel prior、 color line、或是negative correction，根據這些不同的prior information，可估計出不同的transmission map。每一種prior information做出來的方法會影響最後除霧效果的表現。根據每一種prior information會對transmission map中不同的地方產生enhancement，進而達到不同的除霧效果。然而這些演算法主要適用於正常白天環境光源下的霧霾影像，在夜晚環境光源下的影像並不適用。夜霧影像有許多獨特的問題，包括人造光源以及低亮度的影響，會造成正常除霧演算法的失真，故有許多針對處理夜霧影像的方法問世。本文第二個部分主要探討夜間霧霾影像的除霧演算法，探討的方法主要有color transfer method、new imaging modeling、 glowing effect removal，以及 image fusion based 等方法，其主要演算法設計的概念為將人造光源所造成的霧霾影像的失真給去除，使得夜霧影像可以接近白天霧霾的影像，進而使用一般的除霧演算法來處理改變後的夜霧影像。第三個部分則為影像除煙。我們發現在煙霧影像中，在煙霧區域的色彩會有不均勻的失真現象，這是因為在每個color channel的粒子濃度不完全均勻分布所造成，針對此種現象，每一個color channel需要不同的transmission map來做處理。因此，我們提出single dark channel prior的觀念使用dark channel prior的除霧演算法來做影像除煙，更利用多次除霧的過程，使每一次除完的煙霧殘留更加稀少。此外，透過夜晚影像強化的演算法，強化夜晚有煙霧的影像，再使用本文提出的煙霧消除演算法，亦可在夜晚影像中去除部分煙霧。我們相信影像除煙會是在除霧演算法逐步成熟之後，下一個需要被進一步研究的課題。

關鍵字

影像除霧 ； 夜霧影像除霧 ； 影像除煙 ； 影像復原

並列摘要

This thesis discusses about how hazy imaging model can be applied in many fields such as daytime dehazing, nighttime dehazing or moreover, image desmoking. The first part of the thesis is about some important existing daytime dehazing algorithm such as dark channel prior method, color line method, or negative correction model. These methods take different prior information to recover the non-hazy scene. According to these different priors, we can acquire different transmission maps and recovered results. Different priors will enhance different part of hazy images depending on the original assumption of priors and lead to different performance of different dehazing algorithm. However, these algorithms are just suitable for daytime hazy images and cannot be applied on nighttime hazy images. Nighttime hazy images usually contain artificial light source and have low luminance compared with daytime hazy images. In our second part, we introduce different nighttime dehazing algorithms including color transfer method, new imaging modeling, glowing effect removal, and image fusion based method. These algorithms pre-process nighttime hazy image to make them look like daytime hazy images and use existing daytime dehazing algorithm to recover nighttime non-hazy scenes. In our third part, we propose a novel desmoking algorithm based on hazy imaging model. We discover that there is hue distortion in smoke region of smoky images which results from unbalanced particle density distributed in each color channel. In our methods, we propose single channel dehazing based on dark channel methods to make different transmission map of each color channel. Moreover, we iteratively dehaze different color channels to make residual smoke less and less. Inspired by nighttime image dehazing, we observe nighttime smoky images share the same problem with nighttime hazy images including low lighting condition and multiple scattering process between artificial light source and particles. We propose nighttime image desmoking algorithm based on our proposed daytime desmoking algorithm. Night vision enhancement preprocessing is applied to nighttime smoky image to change it to daytime-like. Proposed daytime desmoking algorithm is then applied on enhanced images. Results shows that detail in both low luminance region and smoke region is enhanced.

並列關鍵字

image dehazing ； nighttime haze removal ； image desmoking ； image restoration

參考文獻

[35] Narasimhan, Srinivasa G., and Shree K. Nayar. "Vision and the atmosphere." International Journal of Computer Vision 48.3 (2002): 233-254.

[2] Fattal, Raanan. "Single image dehazing." ACM transactions on graphics (TOG) 27.3 (2008): 72.

[3] He, Kaiming, Jian Sun, and Xiaoou Tang. "Single image haze removal using dark channel prior." IEEE transactions on pattern analysis and machine intelligence 33.12 (2011): 2341-2353.

[4] Gao, Yuanyuan, et al. "A fast image dehazing algorithm based on negative correction." Signal Processing 103 (2014): 380-398.

[5] Gibson, Kristofor B., and Truong Q. Nguyen. "An analysis of single image defogging methods using a color ellipsoid framework." EURASIP Journal on Image and Video Processing 2013.1 (2013): 37.