請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/58936
標題: | 對多物件去模糊之立體視覺演算法與硬體設計 A Stereo Vision Algorithm and its Hardware Design for Multiple Object Deblurring |
作者: | Yang-Yao Lin 林暘曜 |
指導教授: | 盧奕璋(Yi-Chang Lu) |
關鍵字: | 去模糊,深度估測,點函數分布, Deblurring,Depth estimation,Point spread function, |
出版年 : | 2013 |
學位: | 碩士 |
摘要: | 在影像去模糊處理的研究中,如果模糊核具有平移不變性和均勻的狀況下,空間域的去模糊技術通常是一個有效的方法。然而,由於遮蔽和非剛性的扭曲,使得運動模糊往往是不具有平移不變性而且不均勻的。在這篇論文裡,我們提出了利用時域點分布函數對多物體與模糊核在空間域中不具平移不變性的場景進行去模糊的方法。透過控制相機的曝光時間使時域點分布函數具有平移不變性是本論文一個重要的優勢。
在我們的方法中,利用立體相機,可以得到不同曝光時間拍出的左圖與右圖。我們先利用左右圖產生出一個粗略的深度圖,再利用深度資訊進行去模糊。因為這些圖片是同時被拍攝的,兩張圖之間不會有延遲,所以真實景物的細節是可以被還原出來而不需要有關於景物的預先知識。此外,大多數的相機為避免模糊選擇了短曝光時間,在此情況下很難取得完整的色彩資訊,但是我們的方法能透過套用原圖的真實色彩而不會失去色彩資訊。 在本篇論文中實作了深度估算電路和去模糊電路的硬體,能夠在22毫秒內完成影像大小為 的去模糊運算,達到即時(Real -time)處理的標準。我們使用TSMC 90nm製程,深度估算電路與去模糊電路之面積分別為5.5 μm2和0.6 μm2,操作時脈為100 MHz。 Spatial deblurring techniques are reasonably well with an assumption that the blur kernel is shift-invariant and uniform. However, motion blurs are often shift-variant and non-uniform due to motion occlusions and non-rigid distortions in the scene. In this thesis, we propose a general deblurring approach for multiple objects and spatial shift-variant blur kernels using a temporal point spread function (PSF). An important advantage of adopting the temporal blur model is that the temporal PSF, unlike spatial PSF, can be shift invariant by controlling the exposure time. In our approach, with a stereo camera, both left scene and right scene are captured with different exposure time. We first generate a rough depth map from the two scenes, and then perform deblurring differently according to object depths. Since the scenes are captured simultaneously, there is no delay information between two images so that the detail of the real scene can be reconstructed without image priors. In addition, sometimes cameras cannot obtain sufficient color information when short exposure time is used to avoid motion blur. Our method records color information of blurred images and applies the information to the deblurred image. Consequently, true color information can be maintained on our deblurred images. In this thesis, we implement a depth estimation circuit and a deblurring circuit to reduce processing time. The two chips are implemented using TSMC 90 nm cell library. Our chips can finish deblurring within 22 ms for a image, thus real-time processing capability can be achieved. The chip areas are 5.5 μm2 and 0.6 μm2, respectively. Both chips operate at 100 MHz. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/58936 |
全文授權: | 有償授權 |
顯示於系所單位: | 電子工程學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-102-1.pdf 目前未授權公開取用 | 6.64 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。