視頻圖像去霧霾算法研究與實現

孫光民 趙瑩帝

摘要 通過課題的研究背景及意義，可以明確視頻圖像去霧霾的必要性。因為圖像增強技術去霧霾時只增強對比度，去霧霾效果不好，所以本文選用圖像復原算法。首先，結合暗通道先驗理論和大氣散射模型，實現單幀圖像去霧霾。該方法會導致天空等特殊區域出現失真。然后，通過閾值法、定值法和容差機制法對大氣光系數和透射率做出改進，總結出五種去霧霾優化算法。再結合調查問卷的結果從主觀評價去霧霾效果，記錄算法運行時間從客觀評價去霧霾速度。最后，對背景靜態的視頻進行去霧霾。先提取背景幀，然后求取通用大氣光系數和通用透射率，再用它們給每一幀圖像去霧霾。此方法移除霧霾后視頻不同幀圖像之間無色差，同時去霧霾速度比較理想。

關鍵詞：暗通道先驗;圖像去霧霾;視頻去霧霾;容差機制

中圖分類號： TP39 ? ? ? ?文獻標識碼：A

文章編號：1009-3044（2020）17-0167-03

Abstract： Through the research background and significance of the topic， it is possible to clarify the necessity of removing haze in video images. Because the image enhancement technology only enhances the contrast when haze is removed， the effect of defogging is not good， so the image restoration algorithm is used in this thesis. Firstly， the single-frame image defogging is achieved by combining the dark channel prior theory and the atmospheric scattering model. This method will cause distortion in special areas such as the sky. Then， the atmospheric optical coefficient and transmittance are improved by the threshold method， fixed value method and tolerance mechanism method， and five kinds of dehaze optimization algorithms are summarized. Combined with the results of the survey questionnaire from the subjective evaluation to the haze effect， the recording algorithm running time objectively evaluates the haze speed. Finally， go to the background static video to go haze. The background frame is extracted first， then the general atmospheric light coefficient and the universal transmittance are obtained， and then they are used to de- haze each frame of image. This method has no color difference between the images of different frames of the video after the fog is removed， and at the same time， the speed of de- haze is ideal.

Key words： dark channel prior;image defogging;video defogging;tolerance mechanism

1 緒論

視頻圖像在我們的生活中被廣泛應用。霧霾影響視頻圖像的拍攝效果，妨礙我們直接從中直接獲取有用信息。這說明視頻圖像去霧霾很有必要。目前視頻圖像去霧霾軟件技術主要分為圖像增強算法和圖像復原算法。圖像增強算法只注重對比度的提高，有時會出現殘霧。圖像復原算法從成像原理上進行去霧霾。最終本文使用圖像復原算法分支下基于先驗信息的暗通道先驗理論進行視頻圖像去霧霾。

2 基于暗通道先驗理論圖像去霧霾的實現

暗通道先驗理論[1]最簡潔的描述：在絕大多數非天空的區域里，某一些像素總會存在至少一個通道具有很低的值。……