使用關鍵通道縮放與環境光夾角調整之快速單張影像去霧技術

Abstract

霧是一種常見會影響能見度及對比度的大氣現象，其原因來自於大氣粒子散射環境光造成，其影響程度和物體與相機接收端之距離呈正相關。單張影像去霧因為其輸入資訊較少，在影像處理中屬於極富挑戰性的非良置性問題，也因此近年來有大量研究投入。本論文提出新式的單張影像去霧演算法屬於穿透率去霧演算法，使用顏色在RGB空間中之夾角找出環境光，再藉由環境光夾角調整關鍵通道縮放的概念反推穿透率，並進一步使用引導影像濾波器將穿透率加入空間平滑性。除了影像去霧外，本論文也針對穿透率去霧結果進行後處理，包含了亮度修正、顏色突出兩個部分，對於去霧後的亮度和色彩分別進行補償以得到品質更好的影像。 相關研究往往會有去霧影像過暗及運算時間過長等問題，本論文提出之全新單張影像去霧演算法去霧效果佳，不會使去霧影像過暗；且運算時間相當快速，適合應用在即時影片去霧，提升行車安全。提出之後處理演算法通用性相當高，可以銜接在相關穿透率去霧研究以調整去霧結果。

Haze is a common atmospheric phenomenon that can significantly degrade the visibility and contrast of a scene. The main cause of it is air-light scattering by the particles in the atmosphere. The impact of haze and the depth of the object are positively correlated. Single image dehazing is a challenging ill-posed problem in image processing because of the insufficiency of input information. Therefore, there were many researches focused on this issue these years. In this thesis, we propose a fast and novel transmission-based single image dehazing algorithm, using the angles of colors in RGB space to estimate air-light. Then we scale the critical channel value of the image by considering the angles between the image color and air-light color. Then we can use the value of the critical channel to estimate the transmission and refine it by applying a guided image filter. We also propose luminance correction and stand-out color enhancement algorithm for post-processing of image dehazing, in order to deal with over-dehazing and low contrast problem in dehazed image, respectively. The related works are often suffered from either over-dehazing issue or lengthy computation time. The proposed dehazing algorithm has well dehazing result without over-dehazed effect. Besides, the computation time is short so that it is ideal for video dehazing. The proposed post-processing algorithms can adjust the dehazed results and they are both suited for transmission-based image dehazing algorithms.