改良之暗通道先驗除霧方法

A Modified DCP Based Dehazing Algorithm

陳俊佑(Chun-Yu Chen)

指導教授 ： 謝政勳

朝陽科技大學/資訊學院/資訊工程系/碩士(2019年)

摘要

在不佳的天氣條件下拍攝影像，例如煙、霧、下雨或下雪等，通常會對拍攝影像造成不良的影響。在有霧的情況下，拍攝的影像往往呈現灰白的現象，影響影像的對比度和能見度，為了解決霧氣對影像的影響，許多學者專家就提出了除霧演算法。在近來的除霧演算法中，基於暗通道先驗(Dark Channel Prior, DCP)的方法受到廣泛的注意和研究。儘管在一般情況下，傳統的DCP方法具有令人滿意的效果，然而，DCP方法仍然存在著三個問題：第一、在天空區域會發生人工瑕疵；第二、在景深不連續處會發生光暈；第三、需要較高的計算成本。為了解決上述DCP方法的問題，本論文提出了一種改良的DCP方法(Modified DCP, MDCP)。在MDCP方法中，我們修改了傳統DCP方法估計大氣光和透射率的方式，在大氣光估計中，我們使用基於像素的暗通道和自適應的調整因子。在透射率估計中，採用自適應調整因子取代固定調整因子來估計初始透射率。此外，像素暗通道用於導引影像濾波器中的導引影像，並且採用較大的濾波視窗和平滑因子，為了驗證本論文MDCP方法的可行性，我們以多張影像作為範例，並利用主觀視覺品質與客觀除霧性能指標來評估性能。此外，MDCP方法也與傳統DCP方法以及其他除霧方法進行比較，結果顯示，所提出的MDCP方法恢復影像的主觀視覺舉客觀除霧性能指標通常都有較佳的結果。
關鍵詞：暗通道先驗、導引影像濾波器、透射率估計。

關鍵字

暗通道先驗 ； 導引影像濾波器 ； 透射率估計

並列摘要

Taking an image under adversary weather conditions, such as smoky, foggy, rainy or snowy conditions, will degrade the image quality generally. In a hazy condition, the image would suffer from low contrast and poor visibility. To rid of the problem, haze removal algorithms are sought by researchers. Recently, an algorithm based on dark channel prior (DCP) has been attracted much attention. In spite of the conventional DCP algorithm usually has satisfactory restored images, at least problems have to be solved further. First, artifacts may occur, especially in the sky region. Second, halos may happen at large depth discontinuities. Third, high computational cost is required. To relieve the problems, this thesis proposes a modified DCP (MDCP) algorithm where the way to estimate atmospheric light and transmission is modified. In the atmospheric light estimation, the pixel-based dark channel is employed with an adaptive factor. In the initial transmission estimation, an adaptive factor is used to replace the fixed factor in the DCP algorithm. To refine the initial transmission, the setting for the guide image filtering is changed where the pixel-base dark channel is employed as guide image and large window size and smoothing factor are used. To verify the proposed MDCP algorithm, several examples are given. Moreover, the proposed MDCP algorithm is compared with the conventional DCP and other recently reported algorithms. The results indicate that the proposed MDCP algorithm generally outperforms the compared algorithms in terms of subjective visual quality and objective dehazing performance index.
Keywords：: dark channel prior, guide image filtering, transmission map estimation

並列關鍵字

Dark Channel Prior ； Guide Image ； Transmission Map Estimation

參考文獻

[1] K. He, J. Sun, and X. Tang, "Single image haze removal using dark channel prior," IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 33, Issue 12, pp. 2341-2353, 2011.

[2] Wencheng Wang, Xiaohui Yuan, Xiaojin Wu, Yunlong Liu, Somayeh Ghanbarzadeh, "An Efficient Method for Image Dehazing," IEEE International Conference on Image Processing, pp.2241-2245, 2016.

[3] Elisee A Kponou, Zhengning Wang, Ping wei, "Efficient Real-time Single Image Dehazing Based on Color Cube Constraint," IEEE International Conference on Signal and Image Processing, pp. 106-110, 2017.

[4] Yuhei Kudo, Akira Kubota, "Image Dehazing Method by Fusing Weighted Near-Infrared Image," International Workshop on Advanced Image Technology, pp. 1-2, 2018.

[5] Qi Liu, Xinbo Gao, Lihuo He, Wen Lu , "Single Image Dehazing With Depth-Aware Non-Local Total Variation Regularization," IEEE Transactions on Image Processing, Vol.27 , Issue.10, pp. 5178 – 5191, 2018.