以視網膜-大腦皮質為基礎的智慧型影像處理技術

Retinex-Based Intelligent Image Enhancement Technologies

連蕙君(Hui-Chun Lien)

指導教授 ： 貝蘇章

國立臺灣大學/電機資訊學院/電信工程學研究所/碩士(2008年)

https://doi.org/10.6342/NTU.2008.01480

摘要

現今數位相機幾乎是每人必備的生活紀錄工具，但市面上的相機所拍攝的影像，卻時常跟我們眼睛所見到的不同。這些影像是可以由後製處理使其更美觀、清晰或接近真實所見，可是這通常需要人為的判斷與調整。如果要處理的影像數量龐大，或是想要讓一般不會影像後製的民眾也能獲得美觀的照片，亦或者希望把處理內建在相機中，都需要自動影像美化的技術。這篇論文的目標就是完成一個全自動的影像美化系統。
影像美化的方法主要就是調整亮度、對比，還有色彩的平衡。傳統自動影像美化的技術多半是整張影像作調整，容易有不自然的效果；而在這裡我們引進視網膜-大腦皮質(Retinex)技術，可以針對影像內容的不同，作局部的調整。我們在論文中實作並比較了一些研究提出的方法，整合各項技術的優點，經過一些測試圖片決定使用參數，最終建立一個全自動的系統。不需要人工手動調整，即可處理任何影像，達到暗部、亮部皆清晰且自然的結果，並解決強光下的陰影中細節不清晰，或濃霧中影像對比過低的問題。我們的自動影像美化技術相較於一些傳統技術，能夠讓各類型的影像皆能達到自然美觀的結果，局部的對比與整體的色調都有不錯的表現。這篇論文的研究讓全自動美化系統不再是遙不可及的夢想。

關鍵字

人類視覺 ； 動態範圍壓縮 ； 多規模視網膜-大腦皮質技術 ； 自動影像美化

並列摘要

Nowadays digital camera is a necessary tool for people to record their daily life, but images taken by current cameras are usually different from the realistic view. We can use some image post-process techniques to get more pleasant, clearer image or make it nearer to the realistic view, but these techniques need manual adjustment. If there are a large amount of images need to be enhanced, then manual adjustment is not practical; another truth is that not everyone can do image post-process. Thus an automatic image enhancement technique can be very helpful, especially an embedded automatic processor for camera. Our goal is to develop an automatic image enhanced system.
Basic image enhancement is accomplished by adjusting the intensity, the contrast, and the color balance. The conventional automatic image enhancements use global adjustment. Here, we introduce Retinex method to perform the local image enhancement. It can adjust images locally depending on the content of the image. We implement and compare several published algorithms. By combining the advantages of these algorithms and determining the parameters by testing them on known images, we implement an automatic system successfully. It can enhance images without any manual control or adjustment. It is robust enough to deal with almost any kind of images. Those fine details in dark and bright region are both revealed clearly after process. The unclear part in the shadow of sunlight and the low contrast of foggy image can be resolved by our system. It can deal with much more varieties of images as compared to the conventional techniques. It also gives a pleasant, nature image. The local contrast and tonal rendition are both quiet good. This study shows that automatic image enhanced system can be a reality and no longer be an unreachable dream.

並列關鍵字

Human vision ； dynamic range compression ； multi-scale retinex (MSR) ； automatic image enhancement

參考文獻

Chapter 2 Human Visual Perception

[1] E. Land and J. J. McCann, “Lightness and Retinex theory,” J. Opt. Soc. Amer. A, vol. 61, pp. 1–11, 1971.

[2] E. Land, “Recent advances in Retinex theory and some implications for cortical computations: Color vision and the natural image,” Proc. Nat. Acad. Sci., vol. 80, pp. 5163–5169, 1983.

[3] E. Land, “An alternative technique for the computation of the designator in the retinex theory of color vision,” in Proc. Nat. Acad. Sci., vol. 83, pp. 3078–3080, 1986.

[5] A. Rizzi, D. Marini, L. Rovati and F. Docchio, “Unsupervised corrections of unknown chromatic dominants using a Brownian-path-based Retinex algorithm,” J. Electronic Imaging, Vol. 12, Issue 3, pp. 431-441, July 2003.