在這篇論文中,我們提出了一個以機器學習為基礎的影像還原演算法,還原線性運動模糊之數位影像。首先,我們透過強健整體運動估測的結果來初始估計動態模糊參數,並且提出一個新的系統架構,藉由遞迴地調整還原所使用的動態模糊參數來得到更好的影像還原結果。此外,我們也對遞迴調整過後的動態模糊參數在時間軸上進行濾波,以整合真實影片在時間軸上所提供的資訊。最後,我們應用一些後處理的技巧,包含 histogram equalization 以及 bilateral filtering 來改善影像還原的結果。值得注意的是,我們藉著 Support Vector Regression (SVR)的技術,由包含不同模糊程度的訓練影像中,學習出一個無參考影像之影像品質衡量模組。透過實驗結果的呈現,包含對影像品質衡量的測試,以及對模擬影像及真實模糊影片所做的影像還原結果,我們映證了本篇論文所提之動態模糊影像還原演算法的功效。
In this thesis, we propose a learning-based image restoration algorithm for restoring images degraded by linear motion blurs. The motion blur parameters are first approximately estimated from the robust global motion estimation result. Then, we present a novel framework to refine the image restoration iteratively based on recursively adjusting the motion blur parameters for image restoration to achieve the best image quality measure. The temporal information from the frame sequence is also integrated by temporally filtering the refined motion blur parameters for the whole sequence. Finally, we apply some post-processing skills, including the histogram equalization and the bilateral filtering, to improve the image deblurring results. Note that a no-reference image quality assessment model is learned by training Support Vector Regression (SVR) from a collection of representative training images simulated by degradation and restoration with different combinations of motion blur parameter values. Some experimental results on the deblurring of simulated blurred images and real videos are given to demonstrate the performance of the proposed blind motion deblurring algorithm.