單一影像除霧、去雨/雪及水下強化之數位影像技術

Single Image Dehazing, Rain/Snow Removal and Underwater Enhancement Using Digital Image Processing

蔡雨泰(Yu-Tai Tsai)

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(2014年)

https://doi.org/10.6342/NTU.2014.00358

摘要

在惡劣環境下所造成的低能見度，是許多電腦視覺應用的主要問題，如監視，智能車輛，室外物體識別等。這是由於大氣微粒大量存在於參與介質(participating medium)中並且有顯著的分佈和大小時，所造成的現象。基於這一現象，可以將氣候狀況分為靜態和動態兩類。靜態的惡劣氣候是由很多微小的粒子所造成的，其通常在時間和空間上是一致的,如霧和霾；相反地，動態的惡劣氣候是由較大的粒子所造成的，如雨滴和雪花，其在分析上較靜態惡劣氣候困難，主要是因為被雨和雪影響的鄰近區域，在時間和空間上是不同的。然而在水下攝影所造成的能低能見度，最主要的原因是由於色偏及色散現象所造成的。色偏現象為光線於水中傳播時，因不同波長具相異之能量衰減程度，而令水下環境呈現偏藍色調；色散現象則為物體反射光線經水中粒子吸收與多次漫射所造成的，色散現象會對影像產生能見度與對比降低的影像。這些狀況不僅會困擾和混淆人類的觀察者，還會降低那些依賴微小特徵的電腦視覺演算法之有效性，因此模擬在各種環境下所造成的視覺影響和發展出一套良好的演算法來移除與消除這些由懸浮微粒及光線衰減所造成的影響，變得不可或缺與重要的！
在本篇論文中，我們首先介紹目前最具代表性的三種單一影像除霧的方法：contrast-based, independent component analysis, 和dark channel prior-based，並且基於dark channel prior-based的方法，我們更進一步的提出了一個有效且健全的方法來改善除霧的品質。相較於目前存在的方法，我們的方法除了可以對白天霧影像給予滿意的除霧品質，甚至對於夜晚霧影像，仍然可以提供良好的除霧結果。接著，我們介紹目前具代表性的四種去雨及除雪的方法：guidance image based, image decomposition analysis, adaptive nonlocal means filter, 和frequency-based analysis。接著，我們另外提出了一個簡單但有效的去雨及除雪的方法，其主要的設計概念為將去雨及除雪的架構分成兩個部分，第一個部分為偵測及第二個部分為修補。除此之外，我們還介紹目前最具代表性的三種水下影像強化的方法： histogram-based equalization, wavelength-based compensation, 和fusion based。接著，我們提出了一個簡單但有效的水像影像強化方法，其主要的設計概念為結合色彩校正、對比度擴展及直方圖均化的方法，相較於目前存在的方法，是一個可以用較少的運算時間且可有效提高影像能見度與色彩保真度的方法，相信在優化的情況下，能在硬體上達到即時影像增強。

關鍵字

除霧 ； 除霾 ； 除雨 ； 除雪 ； 水下影像 ； 影像復原 ； 視頻復原 ； 影像增強

並列摘要

Poor visibility in bad environment is a major problem for many applications of computer vision such as surveillance, intelligent vehicles, and outdoor object recognition, etc…. The reason is that the substantial presence of atmospheric particles has significant size and distribution in the participating medium. Based on this, weather conditions can be characterized as steady and dynamic cases. Specifically, steady bad weather such as fog and haze caused by microscopic particles is usually spatially and temporally consistent. Oppositely, dynamic bad weather such as rain and snow in made up of large particles. Because spatially and temporally neighboring areas are affected by rain and snow differently, the analysis is more difficult. However, the poor visibility in underwater photography is caused by light scattering and color shift. Color shift corresponds to the varying degrees of attenuation encountered by light traveling in the water with different wavelengths, rendering ambient underwater environments dominated by bluish tone. Light scattering is caused by light incident on objects reflected and deflected multiple times by particles present in the water before reaching the camera. This in turn lowers the visibility and contrast of the image captured. Under these conditions, the human viewer would be annoyed. They also degrade the effectiveness of any computer vision algorithm based on small features and varying degrees of attenuation. Therefore, it is necessary to model the visual effects for the various cases and then remove them.
In this thesis, we introduce three existing typical single image dehazing methods: contrast-based, independent component analysis, and dark channel prior-based. To improve the dehazing quality, we propose a robust and effective dehazing method. Unlike other existing methods, there is the satisfactory dehazing quality during daytime and nighttime by our methods. And then, four existing typical rain and snow removal methods in single image: guidance image based image decomposition analysis, adaptive nonlocal means filter, and frequency-based analysis are also introduced in the literature. In this follows, we design a simple but effective method divide the rain or snow removal scheme into two parts, the first part is detection of rain or snow and the second part is inpainting. Besides, three existing typical underwater enhanced methods: histogram-based equalization, wavelength-based compensation, and fusion based are also introduced in the literature. In this follows, we design a simple but effective underwater enhanced method, and its main idea is combining the color correction, contrast stretching, and histogram equalization. Unlike other existing methods, we’ll get a better result which takes less processing time and highly enhances visibility and superior color fidelity by our method. We believe that we’ll run real-time on hardware in optimized circumstances.

並列關鍵字

Dehazing ； Fog removal ； Rain removal ； Snow removal ； Underwater Image ； Image Restoration ； Video Restoration ； Image Enhancement

參考文獻

[2] R. Fattal, “Single Image Dehazing,” ACM Transactions on Graphics, vol. 27, pp. 72:1-72:9, August 2008.

[4] J. Xu, W. Zhao, P. Liu, and X. Tang, “An Improved Guidance Image Based Method to Remove Rain and Snow in a Single Image,” CCSE Computer and Information Science Journal 5(3), 2012, pp. 49-55.

[5] L.-W. Kang, C.-W. Lin, and Y.-H. Fu, “Automatic single-frame-based rain streak removal via image decomposition,” IEEE Trans. Image Processing, vol. 21, no. 4, pp. 1742-1755, Apr. 2012.

[9] J.-Y. Chiang and Y.-C. Chen, “Underwater Image Enhancement by Wavelength Compensation and Dehazing,” IEEE Transactions on Image Processing, vol.21, no.4, pp.1756-1769, April 2012.

[11] S. G. Narasimhan and S. K. Nayar, “Vision and the atmosphere,” International Journal of Computer Vision, vol. 48, pp. 233-254, July-August 2002.