Restauration Des Images Et Optimisation Des Poids
Qiyu Jin
2012-01-01
Abstract:In this these, we present and study some methods for restoring the original image from an image degraded by additive noise or blurring. The these is composed of 4 parts,and 6 chapters. Part I. In this part, we deal with the additive Gaussian white noise. This part is divided into two chapters : Chapter 1 and Chapter 2. In Chapter 1, a new image denoisingalgorithm to deal with the additive Gaussian white noise model is given. Like the nonlocal means method, we propose the filter which is based on the weighted average of the observations in a neighborhood, with weights depending on the similarity of local patches. But in contrast to the non-local means filter, instead of using a fixed Gaussian kernel, we propose to choose the weights by minimizing a tight upper bound of mean square error. This approach makes it possible to define the weights adapted to the function at hand, mimicking the weights of the oracle filter. Under some regularity conditions on the target image, we show that the obtained estimator converges at the usual optimal rate. The proposed algorithm is parameter free in the sense that it automatically calculates the bandwidth of the smoothing kernel. In Chapter 2, we employ the techniques of oracle estimation to explain how to determine he widths of the similarity patch and of the search window. We justify the Non-Local Means algorithm by proving its convergence under some regularity conditions on the target image. Part II. This part has two chapters. In Chapter 3, we propose an image denoising algorithm for the data contaminated by the Poisson noise. Our key innovations are : Firstly, we handle the Poisson noise directly, i. E. We do not transform Poisson distributed noise into a Gaussian distributed one with constant variance ; secondly, we get the weights of estimates by minimizing the a tight upper bound of the mean squared error. In Chapter 4, we adapt the Non-Local means to restore the images contaminated by the Poisson noise. Part III. This part is dedicated to the problem of reconstructing the image from data contaminated by mixture of Gaussian and impulse noises (Chapter 5). In Chapter 5, we modify the Rank-Ordered Absolute differences (ROAD) statistic in order to detect more effectively the impulse noise in the mixture of impulse and Gaussian noises. Combining it with the method of Optimal Weights, we obtain a new method to deal with the mixed noise, called Optimal Weights Mixed Filter. Part IV. The final part focuses on inverse problems of the flou images, and it includes Chapter 6. This chapter provides a new algorithm for solving inverse problems, based on the minimization of the L2 norm and on the control of the total variation. It consists in relaxing the role of the total variation in the classical total variation minimization approach, which permits us to get better approximation to the inverse problems. Theix numerical results on the deconvolution problem show that our method outperforms some previous ones