顏嗣鈞臺灣大學:電機工程學研究所於積瑞Yu, Jee-RayJee-RayYu2010-07-012018-07-062010-07-012018-07-062008U0001-2707200814051000http://ntur.lib.ntu.edu.tw//handle/246246/187965在這篇論文中，我們提出了一個直接由模糊的彩色影像中，估計模糊核心的大小和角度的演算法，配合使用已有的去迴旋積演算法即構成一個盲目去迴旋積的架構。在演算法中，我們使用顏色資訊及影像的統計特性來做為估計模糊核心的依據。實驗顯示我們的演算法在個人電腦上能在可被接受的時間之內完成影像品質的改善。We provide an algorithm used to estimate the length and direction of a motion blur kernel directly from the observed color image. The result of estimation can be used in standard deconvolution algorithms, and the whole approach makes a blind deconvolution scheme. The information of different color channels and the gradient statistics of image are used to estimate the blur kernel in our algorithm. Experimental results show that the scheme can reduce the blurring in images in acceptable time on PC.1 Introduction 1 Preliminaries 7 2.1 Image Model 7 2.2 Blur Kernel Model 9 2.3 Deconvolution Methods 10 Estimation of Blur Size 12 3.1 Color Stair Phenomenon 12 3.2 Analysis of Color Stair 15 3.3 Color Stair in Natural Images 18 3.3 Size Estimation Algorithm 20 Estimation of Blur Direction 25 4.1 Gradient Statistics of Nature Image 25 4.2 Direction Estimation Using Gradient Statistics 28 4.3 Direction Estimation Algorithm 30 Experimental Results 35 Conclusion and Discussion 42eference 432731026 bytesapplication/pdfen-US盲目去迴旋積手震blind devonvolutioncamera shakethesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/187965/1/ntu-97-R95921084-1.pdf