請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/15362
標題: | 煙霧瀰漫,沙塵暴與水下環境的影像增強及軟組織去除的醫學應用 Haze, Fog, Sand Removal and Attenuation Effect Compensation for Natural and Underwater Image Enhancement and Medical Image, Soft-Tissue Removal Application |
作者: | Yun-Han Chiu 邱韵涵 |
指導教授: | 貝蘇章(Soo-Chang Pei) |
關鍵字: | 色彩恆常性,低亮度影像增強,影像除霧,白天/夜晚影像除煙,水下影像增強,影像除沙,醫學影像處理, Color constancy,Low light enhancement,Dehazing,Daytime/Nighttime smoke removal,Underwater image enhancement,Sand removal,Medical image processing, |
出版年 : | 2020 |
學位: | 碩士 |
摘要: | 在惡劣環境下,例如霧,煙,沙子和水下的圖像,通常會出現低對比度和低能見度的問題,這些狀況嚴重影響需要高品質圖像的電腦視覺應用。由於空氣中大大小小的懸浮粒子會吸收光的能量並散射光線,減弱了相機拍攝的影像,進而降低了能見度。具體來說,霧的影像通常是分佈均勻且不會隨著空間改變,然而煙的出現會分佈在特定區塊且在空間中的分佈不均勻。在夜間情況下,還會出現色彩偏移以及照明不足的問題,這使得分析上更加困難。對於在水下影像及沙塵暴影像的低能見度,主要原因是色偏及色散,光線傳播的衰減度是取決於光的波長以及環境,在水中紅光會最先被吸收,而在沙塵暴中黃光會散射得最嚴重,因此導致水下影像呈現偏藍綠,而沙的影像偏黃的外觀。在這些情況下,不僅會干擾觀察者,還會使得電腦視覺和一些應用程序(例如監視,智能車輛,物件識別,偵測和追蹤)的有效性也大幅降低。因此模擬在各種環境下造成的視覺影響並設計演算法來移除這些問題是很重要的任務。
在本篇論文中,我們介紹了大氣散射的形成模型,並說明結合人類視覺系統如何應用於不同的環境。首先,我們提出了顏色恆定性的演算法及三種低亮度增強的方法。接著,結合這些方法,我們改良了現有的白天及夜間的除霧演算法。對於煙的影像,根據其分佈及密度不均勻的特性,提出一種局部除煙的演算法,可以保留整體場景的自然性且不失真。此外,我們利用色彩平衡及對比度擴展,設計可以應用於水下影像增強及除沙的有效方法。最後,我們進一步將提出的演算法應用於X射線及視網膜影像。本篇論文成功的提高惡劣環境下的影像能見度並應用於增強醫學影像。 Images taken in bad weather conditions, such as haze, smoke, underwater, and sandstorm, often suffer from low contrast and limited visibility. These scenarios may significantly affect many applications of computer vision, which relied heavily on image qualities. The poor visibility is mainly due to the suspended particles in the air, which absorb and scatter the light, and attenuate the scene radiance captured by the digital camera. Specifically, in the daytime cases, the haze image is usually spatially and temporally consistent, while the smoke image is non-consistent and non-uniform distribution. In nighttime cases, the haze and smoke even suffer from the color shift and low illumination problems, which cause the analysis more difficult. For the underwater and sandstorm images, the attenuation of light depends on the light's wavelength, usually resulting in the green-bluish/yellowish appearance. Under these issues, human perception may be greatly annoyed. The effectiveness of computer vision and consumer applications, such as surveillance, intelligent vehicles, object recognition, detection, and tracking are also degraded. Therefore, it is indispensable to model these effects and eliminate them. In this thesis, we introduce how the atmospheric scattering model and the human visual system can be applied in different scenarios. First, we proposed the color constancy algorithm and three low-light enhancement methods. Then, using these methods, we improve the existing daytime and nighttime dehazing algorithms. For the smoke images, we followed the characteristics of smoke and proposed a local de-smoking algorithm, which can retain the naturalness of the scenes. In addition, we design an effective method that can apply to both underwater and the sandstorm images. In the last part, we show that our proposed method can also be used as a preprocessing step in medical image processing. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/15362 |
DOI: | 10.6342/NTU202001155 |
全文授權: | 未授權 |
顯示於系所單位: | 電信工程學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
U0001-2506202022083100.pdf 目前未授權公開取用 | 9.59 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。