基於最小值圖實現影像與影片去霧

Abstract

這篇論文提供了一個全新的除霧的方法與概念,利用我們所提出 的最小值圖做為依據,對於不同的區塊進行不同程度的除霧,進而達 到最佳的除霧效果。最小值圖代表該點在 RGB 色彩空間中,擁有最 小值的色彩通道 (Channel) 為何,我們發現一張圖擁有許多不同顏色的 區域,而這些區域除了會在最小值圖上呈現外,也會顯現出不同的特 性。基於此觀察,對於不同的區域我們會各自計算屬於該區域的大氣 光 (atmospheric light) 值,利用區域合併、大氣光估算以及平滑化濾波 器來計算出最好的整張圖的最佳大氣光值;而對於大氣幕 (atmospheric veil),我們則是先利用最小值圖求出每個點的最小值後,利用均值濾 波器、雙邊濾波器、引導濾波器,我們將可以獲得適當的大氣幕,在 有了大氣光與大氣幕後,我們可以利用影像霧模型成功地將有霧的影 像還原成沒有霧的真實景。對於影片,我們使用了空間時間濾波器對 影片進行優化,使影片可以同時保有時間上與空間上的一致性。實驗 結果顯示出我們提出的方法,可以有效地對影像實現去霧,與他人的 影像結果相比,我們結果更為自然、色彩飽和,對於細節也有良好的 去霧能力;對於影片也可以達到一致性,輸出高品質的影片。

A novel image and video haze removal method based on minmap is pro- posed in this paper. Minmap is defined as the channel which contains the minimal component of three RGB color channels. Because one image may contain many regions which differ in color, minmap contains different regions which show the different properties. Based on this observation, atmospheric light is estimated by region growing, dark channel prior and smooth filter we proposed separately for each minmap region. To compute veil, we estimate the whiteness based on minmap first. Then, median filter, bilateral filter and guided filter are used on whiteness to get appropriate veil. After atmospheric light and veil are computed, the hazy images can be successfully recovered using haze image model. For video version, space-time filter is proposed to ensure the temporal and spatial coherence. Compared with exiting state of the art methods, our method could have better haze removal effects. The image results demonstrate that the results are vivid and have the properly contrast in all of the regions. The video results show that the video is high quality with temporal and spatial coherence.