ICA Technology And It’s Application In Image Denoising

Independent Component Analysis (ICA) is a new technology. In the past 10 years, ICA has made rapid development and has proposed many efficient ICA algorithm. ICA can restore the original independent components just by observational information which is mixed by the independent componets and in accordance with a few simple assumptions.Using this character of ICA to separate noisy data and image data which is independent of each other, achieve the effect of the image denoising. ICA-based feature extraction is to use its own statistical properties of the data obtained. For different processing objects, different characteristics of the base will be extracted, so that extracted features of the base has a self-adaptive. Using this feature of ICA to extract the feature of the image, transforming the original noisy image data into the new transform domain in which shrinkage the noisy image data to denoising, thus achieve a better effect of image denoising.ICA algorithm is used to denoise the noisy images in the thesis. Generally, ICA algorithm is divided into two kinds of image denoising methods. One is denoising noisy image based on a virtual channel. In this thesis, on the noisy image wavelet decomposition to build up a virtual channel, the virtual channel as an input to separate a noisy image and virtual channel, so denoised mages can be obtained. Do some experiments for the construction of the virtual channel and different noise, and get the parameters of the denoised images and the traditional denoising method and compared with those parameters. The other method is the sparse coding shrinkage denoising. Namely the use of ICA find a suitable base, thus the noisy image be linear transformated, changing to another transform domain in that the transformed data be shrinked so as to achieve image denoising. In this thesis, FastICA algorithm be used to study the images so that get features of the base, taking advantage of features of the base to achieve the image transformation, using the maximum likelihood method to estimate the weight of the noisy component from the images which is obtained after wiener denoising Similarly, the denoised image parameters will be measured and compare with those parameters by the traditional methods of image denoising. Summing up sparse coding image denoising method is suitable to which kind of image.