Study On Nonlinear Filtering Algorithm For Digital Image Processing

The dissertation firstly introduces the development history, the research content and the theories application realm of the digital image processing, then emphasizes to introduce the picture restoration in the image processing research realm. the main calculate ways of noise removal: restoration in the presence of noise only-spatial filtering and periodic noise reduction by frequency domain filtering. Restoration in the presence of noise only-spatial filtering contains: Mean Filter,Order-Statistic Filter,Adaptive Filter. Periodic noise reduction by frequency domain filtering includes: Bandreject Filter,Bandpass Filter,Notch Filter. At thorough research of the past Gaussian noise filterings, a nonlinear filtering algorithm using probability statistic and main texture direction analysis is proposed. This algorithm utilizes Radon transform to determine texture direction probability density distributions of local areas of images and then applies probability statistic model to estimate the middle pixel's gray value according to its neighbour pixels. The performance of the proposed method is evaluated with several sets of images contaminated by pulse noise and Gaussian noise. The experimental results show the superiority of this method including the ability of denoising and preserving edges and details of images especially for images with pulse noise and Gaussian noise. This algorithm makes richly use of local characteristics and details of images, especially unlike some recent algorithms only for removing salt-and pepper impulse or Gaussian noise, it is applicable to images contaminated by any kind of noises. Image filtering algorithm using double noise detector and edge-preserving regularization function is proposed to remove salt-and-pepper noise. The proposed filter has two schemes: noise detector and noise restoration. In the first phase, noise candidates identified as contaminated pixels with the noise detection algorithm of adaptive median filter are judged again by local fuzzy membership function in order to enhance accurate rate of noise detection. In the second phase, a convex objective function composed of (?)₁ data-fidelity term and edge-preserving regularization function is employed to deal with noise candidates. In order to take advantage of local image feature, the input of edge-preserving regularization function is adaptively selected. The image corrupted by noise is restored as the convex objective function gets the minimum in the noise candidate set. Experimental results show that the superiority of this filter in terms of the ability of removing noises and the ability of preserving the partial details of images in comparison with some recent methods especially when the noise level is over 70%.