數位影像愈來愈普及的現在，失真影像復原相對地變得重要。在現有的影像復原方法中，大部分的方法都假設失真過程的資訊是已知的。然而，在實際的情況下，這些資訊大多不可能事先得知。因此，我們必須利用失真影像的特性或統計值來還原影像。

利用部分有關失真過程的資訊，或者在完全沒有任何的事前的資訊情況下，來還原影像的過程，常稱之為盲影像復原。大部分的盲像復原演算法，都是利用疊代的方法，去估計在影像失真模型中的點擴散函數(PSF)及原始影像。然而，在多數這類的方法中，常會有較高的複雜度。在某些情況下，有些演算法甚至會收斂到一個無法預測的結果。因此，增加演算法的穩定性便成為一個重要的議題。

在本篇論文中，我們利用了獨立成分分析(ICA)的觀念來分析影像中的獨立性。在此也利用這個觀念，來估計點擴散函數的參數值與原始影像。實驗結果顯示對於不同的影像，我們所估計到的參數值，會接近於實際的參數值。此外，對於不同類型的影像，我們的演算法也呈現了一定的穩定性。

Digital images are more and more popular now, and the restoration of the degraded images becomes more important. In the existed methods for image restoration, the informations about the degraded process are supposed to be known. However the informations of the degraded image is not always known in practice. Thus we must use the properties or statistics of the degraded images to restore the these images.

The process of image restoration, which uses only partial or no information about the degraded process, are often the so-called blind image deconvolution. Most algorithms of blind image deconvolution use iterative methods to estimate the point spread function (PSF) and the original image of the degraded model. However, most of these methods have heavy computation complexity. Furthermore some algorithms even converge to a unpredicted result in a certain situation. Therefore it is a important issue to increase the stability of restoration algorithms.

In this thesis, we use the concept of independent component analysis (ICA) to analyze the independence of images. We also use this concept to estimate the values of the parameters of PSFs and the original image. Experimental results present that the estimated values of parameters is closed to the real values, and our algorithm is also stable for various images.

[1] A. Hyvrinen and E. Oja. Independent component analysis: Algorithms
and applications. Neural Networkss, 13(4-5):411{430, 2000.
[2] A. K. Katsaggelos. Digital Image Restoration. New York: Springer-
Verlag, 1991.
[3] S. Kirkpatrick, C. D. Gelatt Jr., and M. P. Vecchi. Optimization by
simulated annealing. Science, 220(4598):671{680, 1983.
[4] F. Krahmer, Y. Lin, B. McAdoo, K. Ott, J. Wang, D. Widemannk, and
B.Wohlberg. Blind image deconvolution: motion blur estimation. Tech.
Rep., University of Minnesota, 2006.
[5] D. Kundur and D. Hatzinakos. Blind image deconvolution. IEEE Signal
Processing Mag., page 4364, May 1996.
[6] D. Kundur and D. Hatzinakos. Blind image deconvolution revisited.
IEEE Signal Processing Mag., 13(6):61{63, Jun 1996.
[7] R. L. Lagendijk, A. M. Tekalp, and J. M. Biemond. Maximum likelihood
image and blur identi cation: a unifying approach. Optical Engineering,
29(5):422{435, 1990.
[8] V. Loyev and Y. Yitzhaky. Initialization of iterative parametric algo-
rithms for blind deconvolution of motion-blurred images. Appl. Opt,
45:2444{2452, 2006.
[9] N. Metropolis, A. Rosenbluth, M. Rosenbluth, A. Teller, and E. Teller.
Equation of state calculations by fast computing machines. J. Chem.
Phys, 21(6):1087{1092, 1953.
[10] R. Molina, J. Mateos, and A. K. Katsaggelos. Blind deconvolution using
a variational approach to parameter, image, and blur estimation. IEEE
Transactions on Image Processing, 15(12):3715{3727, 2006.
[11] A. M. Tekalp, H. Kaufman, and J. W. Woods. Identi cation of im-
age and blur parameters for the restoration of noncausal blurss. IEEE
Transactions on Acoustics, Speech, and Signal Processing, 34(4):963{
972, 1986.
[12] Hujun Yin and I. Hussain. Blind source separation and genetic algorithm for image restoration. In Advances in Space Technologies, 2006 International Conference, pages 167-172. Digital Object Identi er 10.1109/ICAST.2006.313821, Sept 2006.
[13] Y. L. You and M. Kaveh. A regularization approach to joint blur identi-
cation and image restoration. IEEE Transactions on Image Processing,
5(3):416{428, 1996.