植基於快速PSF檢索之非線性動態模糊影像重建

Non-linear Motion Blurred Image Reconstruction based on Fast PSF Retrieval

蔡維昌(Wei-Chang Tsai)

指導教授 ： 黃惠俞

國立虎尾科技大學/電機資訊學院/資訊工程研究所/碩士(2013年)

摘要

在日常生活中，不論是使用何種攝影設備在拍攝影像時因為攝影機晃動造成動態模糊的情形經常發生。在影像曝光的過程中相機晃動所造成的模糊影像，這類的晃動經常是非線性的因此稱為非線性動態模糊，動態模糊經常會使影像品質急劇的下降，只要有使用過手持攝影設備的使用者經常會有類似的經驗。基於這點，為了讓因攝影設備震動造成的模糊影像得以重建，進而獲得清晰的影像，將是本論文的主要目標。在去模糊的研究中都將以非線性動態模糊的軌跡透過點擴散函數(point spread function，簡稱PSF)來表示，並且稱之為模糊核心。本論文主要針對由單一模糊核心造成的全域動態模糊的重建處理，其次再利用所提出的方法應用於由多個模糊核心造成的多重模糊影像重建上。動態模糊的影像重建是一個沒有最佳解的問題，在現行模糊核心估測演算法中通常會利用遞迴的方式來估測模糊核心，遞迴的過程將會相當耗費時間。本論文主要以迭代相位恢復算法和標準化稀疏測量為基礎，提出一個快速且有效的模糊核心檢索演算法，藉由核心分群以及模糊核心整合大幅降低核心比對的時間，並找出最佳的模糊核心。經實驗證明本篇所提之方法可有效的降低執行時間及找出最佳的模糊核心，接下來利用反摺積重建出清晰的影像，且不會因為降低執行時間而影響到重建影像的品質。最後透過本論文提出的演算法應用在多重模糊影像的重建。

關鍵字

去模糊 ； 非線性移動模糊,點擴散函數(PSF) ； 影像重建 ； 相位檢索

並列摘要

In everyday life, when people take a photograph by any kinds of cameras, motion blurred image often happens caused by camera shake. Owing to exposures when captured an image within camera shake, it will make a kind of motion blurred image, this kind of motion blurred phenomenon is often a non-liner motion action. Motion blur always causes the degradation of image quality, as long as the users using the hand-held camera often have a similar experience. For this reason, reconstructing a blurred image into a sharp image will be the main objective in this thesis.
In the past studies, nonlinear motion blur will be modeled as point spread function(PSF) which called blur kernel. This thesis aims to the reconstruction of the global motion blur image caused by a single blur kernel. Secondary, the proposed method is further extended to reconstruct a multi-blurred image caused by the multiple kernels. However, reconstruction a motion blurred image is an ill-pose problem. In state-of-the-art motion blurred estimation methods, these algorithms usually use the recursive method to estimate motion blur kernel. But, the recursive process is quite time-consuming. In order to reduce the execution time, based on iterative phase retrieval algorithm and normalized sparsity measure, we propose a fast best kernel retrieval algorithm based on fast point spread function (PSF) involved iterative phase retrieval method and normalized sparsity measure, which can find the best kernel in a short computing time. Experiment results verify that the proposed method can effectively reduce the execution time and obtain the best motion blur kernel and maintain a high quality of image deblurring. Finally, this proposed algorithm also applies to deblur multiple blurred cases. The deblurring results are acceptable.

並列關鍵字

deblurring ； non-linear motion blur ； point spread function (PSF) ； image restoration ； phase retrieval

參考文獻

[1]Y. W. Tai, H. Du, M. S. Brown, and S. Lin, “Correction of spatially varying image and video motion blur using a hybrid camera,” IEEE Trans. on Pattern Analysis and Machine Intelligence, vol. 32, no. 6, pp.1012-1027, 2010.

[3]D. Krishnan, T. Tay, and R. Fergus, “Blind deconvolution using a Normalized Sparsity Measure,” in Proc. of IEEE Conf. on CVPR, pp. 223-240, 2011.

[4]W. Hu, J. Xue, and N. Zheng, “PSF estimation via gradient domain correlation,” IEEE Trans. on Image Processing, vol. 21, no. 1, pp. 386-392, 2012.

[5]H. Takeda and P. Milanfar, “Removing motion blur with space–time processing,” IEEE Trans. on Image Processing, vol. 20, no. 10, pp. 2990-3000, 2011.

[6]H. Ghennioui and F. Bourzeix, “Architecture solution for real-time deblurring image/video technique,” in Proc. of IEEE Conf. on ICMCS, pp. 1-5, 2011.