基於天空偵測與可適性色彩補償之沙塵影像還原研究

Abstract

隨著全球氣候變遷加劇，世界各地極端氣候現象發生的頻率與強度皆有所增加，其中本文的研究主角—沙塵暴也是其中具有代表性的氣候現象之一，對於智慧監控系統、智慧交通系統及各類的戶外電腦視覺應用皆造成嚴重影響。在沙塵暴環境下，大量的沙塵粒子會吸收光線並散射光線，導致影像中顏色偏差、清晰度下降及細節模糊等問題，這些影像品質的劣化，會影響電腦視覺系統在各種應用場景下的準確性與穩定性。 目前對於沙塵暴影像的增強方法主要可分為色偏校正與細節增強兩個方向。前者多以不同色彩空間的轉換與統計方法進行顏色補償與調整，後者則著重於提升影像對比度與去霧處理。 本研究提出了一種基於天空區域偵測的沙塵暴影像增強方法。該方法綜合了色偏校正、對比度增強、天空區域偵測、分割及除霧技術，以達到提升影像品質的目的。實驗結果顯示，本方法不僅能有效提升影像的清晰度與對比度，同時能保持影像的自然性與真實感。相比於傳統的除霧方法與單一的色偏校正方法，我們的方法具有更好的適應性與穩定性，適用於不同場景與不同程度沙塵影像的增強。 總結來說，本研究提出了一個具有高適應性及穩定性的影像增強方法，能在各種沙塵環境下大幅改善影像品質。未來我們將著重於提升方法的計算效率與處理速度，並嘗試將本方法應用於更多的實際應用場景中，如智能交通、無人機監控、以及自動駕駛系統等。

As global climate change intensifies, extreme weather events have become increasingly more often and stronger. Among these phenomena, sandstorms—a central focus of this study—stand out as a representative phenomenon that impact intelligent surveillance systems, smart transportation networks, and many outdoor computer vision applications seriously. In sandstorm conditions, a lot of dust particles absorb and scatter light, resulting in color distortion, clarity reduced, and blurred details in images. These problems make image quality worse and affect the accuracy and stability of vision-based systems in many applications. Current methods for enhancing sandstorm images mainly divide into two categories: color correction and detail enhancement. Color correction usually use different color space transformation to adjust and recover the color. On the other hand, detail enhancement focuses on improving image contrast and removing haze. This study proposes a sandstorm image enhancement method based on sky region detection. The method combines color correction, contrast enhancement, sky detection and dehazing technique to improve the sandstorm image quality. The experiment shows that our method not only improves the clarity and contrast, but also keep the image natural and real. Compare with only using dehazing or only color correction, our method is more stable and suitable for different kinds of sandstorm images. In conclusion, this study presents a method that is stable and adaptive, and can enhance the image quality under sandstorm condition effectively. In the future, we will try to make it faster and more efficient, and hope to use it in real applications, such as smart transportation, drone surveillance, and autonomous driving system.