Research On Image Noise Denoising Method Based On ICA

Independent Compont Analysis(ICA)is signal processing technology which effectively focuses on blind signal separation in recent years. This technology is widely applied in image process, speech recognition and biomedical signal processing fields because it does not need much priori knowledge of the original signals. Currently, ICA has become a hot area in the field of image processing.In practice, some images contain Gaussian noise and ? stable distribution impulse noise, such as CT in biomedical images, which makes images lost important details, and seriously impacts on peoples’ correct analysis of images. Most of the traditional denoising methods aim at only one kind of noise, so it is difficult to achieve the satisfactory results in removing mixed noise. Therefore, this thesis studies on ICA, and proposes corresponding improvements. The main work and innovations are as follows:(1) The traditional image denoising algorithm result decreased significantly, when images containing ? stable distribution impulse noise and Gaussian noise. This thesis presents an algorithm combined fractional lower order FastICA based on negentropy and VNLMS algorithm together, which accomplishes removal of ? stable distribution impulse noise and Gaussian noise. Specifically, fractional lower order FastICA, which is based on negentropy, mainly focuses on denoising ? stable distribution impulse noise. This method strengthens the effectiveness to handle? stable distribution impulse noise of the traditional algorithm, which is based on second order statistics; VNLMS algorithm aims at removing Gaussian noise and it has good effect on Gaussian noise removal. Simulation results show that this algorithm has good characteristics in separating images which have ? stable distribution impulse noise and Gaussian noise, therefore it has practical significance.(2) This thesis presents fractional lower order PSO-ICA with VNLMS algorithm and fractional lower order QPSO-ICA with VNLMS algorithm aimed at de-noising these two kinds of noise, especially when Newton’S iteration method in fractional lower order FastICA algorithm easily increasing algorithm complexity. The simulation experiments show that the fractional lower order PSO-ICA with VNLMS algorithm is better than fractional lower order FastICA with VNLMS algorithm, and fractional lower order QPSO-ICA with VNLMS algorithm significantlyreduces algorithm run time on the premise of ensuring the denoising effect. In order to fully verify the characteristics of PSO-ICA algorithm and QPSO-ICA algorithm, these two algorithms are applied to analyze the images which only contains Gaussian noise. And the experiments show that the denoising result was improved obviously.