基於自我參考實現影像去模糊

A Self-referenced Method for Image Deblurring

曾映傑(Ying-Chieh Tseng)

指導教授 ： 李明穗

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2011年)

https://doi.org/10.6342/NTU.2011.02129

摘要

在這本篇論文中，我們提出一個基於自我參考之影像去模糊的方法。假設一張部分失焦模糊的影像，有著一些重複性的資訊，對於模糊的區域，我們參考一些來自於清楚區域部分的資訊，來達到去模糊的效果。如此一來，模糊的地方從清楚、銳利的區域找到一個回復的方法，基於這樣的想法，我們發展出一個影像去模糊的方法。
首先，我們使用模糊偵測法(blur detection method)來找到影像上任何一點的模糊程度(blurriness)。利用這個模糊程度將影像分為三類；平滑、模糊與銳利。對於銳利的區塊，我們建立起它們與模糊退化後(Degradation)的對應關係。使得模糊的區塊能夠與銳利的區塊作對應。對於每個模糊的區塊，在此張退化後的影像中，套用兩種尋找相似性(similarity)的方法，為每一個需要處理的區塊(patch)建立一個對原圖的權重對應表，當權重越高時，代表越相似。最後，用最後權重最高的銳利區塊來取代模糊的區塊，以完成回復。實驗結果顯示出我們的方法可以有效的實現影像去模糊，發現此種自我參考去模糊的潛在性以及可能性，並且提供了影像去模糊領域中新的想法與更深入的應用。

關鍵字

失焦模糊 ； 影像去模糊 ； 模糊程度偵測 ； 非區域性演算法 ； 結構相似性

並列摘要

In this thesis, we present a self-referenced method for image deblurring. Assume that there are image redundancies in a partially out-of-focus blurred image. For each blurry patch, it references the information from the sharp regions non-locally, and achieves deblurring. Based on this assumption, an image deblurring method is developed.
First, we apply the blur detection method to get the blurriness value of each position. Classify each patch into three types, smooth, blur and sharp by thresholding. In order to enable the blurry patches to match the similarity sharp patch, create the relative degraded patches for every sharp patch. Then, the similarity weight map is constructed for each blurry patch by using two similarity algorithms, non-local mean weight and the structure similarity. That is, the more similar patch with the target patch, the higher weight it has in its position. Finally, the blurry patch is replaced by the highest weight patch among the sharp patches. In this way, every blurry patch is reconstructed by the sharp and similar patches. Then, the deblurring process is done. Our results demonstrate the proposed method has achieved the removal of blurry effect and show the potential of self-referenced-based deblurring method.

並列關鍵字

Image deblurring ； out-of-focus blur ； non-local mean algorithm ； structure similarity ； blur detection

參考文獻

[2] Y. Tai, H. Du and M. Brown, “Image/Video Deblurring using a Hybrid Camera,” CVPR, 2008.

[3] R. Fergus, B. Singh, A. Hertzmann, S. Roweis and W. Freeman, ”Removing Camera Shake from a Single Photograph,” ACM SIGGRAPH, 2006.

[4] Q. Shan , J. Jia and A. Agarwala, “High-quality Motion Deblurring from a Single Image,” ACM SIGGRAPH, 2008.

[5] J. Jia, “Single Image Motion Deblurring Using Transparency,” CVPR, 2007.

[6] L. Yuan, J. Sun, L. Quan and H. Shum, ”Image Deblurring with Blurred/Noisy Image Pairs,” ACM SIGGRAPH, 2007.