The Research Of SAR Image Denoising Based On ICA

SAR is an important aspect of the remote sensing technology which has importantapplication value in military and civilian aspects. Because SAR image is different from theoptical image, the reasearch methods of SAR image denoising is more important. Blind signalprocessing that is developed in recent years is a new signal processing technique. The purposeis separating independent source signals from the observed mixed signal. Blind signalprocessing plays an important role in many fields, such as wireless communications, speechprocessing, image processing, and biomedical etc. The content of this paper is the SAR imagedenoising method based on ICA. The main research works are as follows:1. To do a comparative analysis of several traditional image denoising method.Discussing the advantages and disadvantages of several image denoising methods, such asmean filtering, median filtering, Wiener filtering, wavelet threshold, apply them to the opticalimage and SAR image to do a simulation. The simulation results show that these methodsexists serious defects for SAR image denoising.2. Making a specific analysis for the fast independent component analysis algorithm thatis based the negative entropy maximization. In the study of the optical image denosing, theFastICA method and traditional denosing method are used to simulate. According to thesimulation results, the FastICA algorithm is more better than others methods.3. Modify the improved FastICA algorithm and appliy it to SAR image. In order to meetICA application mode, a new algorithm which is based on two-dimensional lifting waveletand FastICA is proposed to solve the underdetermined ICA problem.Use the two-dimensionallifting wavelet algorithm to construct a virtual noise channel as observed signal, and then usethe improved FastICA algorithm to separate all of the observed signals. The results showt theproposed algorithm achieves a better balance between the denoising and maintaining imagedetails, and confirms the feasibility and effectiveness of the algorithm.