Research On Multiplicative Noise Removal Based On Principal Component Analysis Method

The development of the image processing is inseparable with the successful applications of mathematical theory. Generally image noise includes multiplicative noise and additive noise. Relatively, multiplicative noise is a more complex one. For example, in synthetic aperture radar image, Gamma distribution noise is a typical multiplicative noise. The presence of Gamma noise seriously affects the quality of the image and hinders the further research of SAR images. Therefore, how to effectively suppress the multiplicative noise has become a key problem in image processing.In this thesis we present an efficient Principal component analyse (PCA) based noise removal method with local pixel grouping. We discussed the mathematical elements of PCA and the PCA-based method for removing additive noise. Further, we propose a new method composed of several stages. First, we change the image from the original field to the Log field using logarithmic function which makes the multiplicative noise to be transformed into additive one. The noise now is studied as additive noise in the Log domain. In additions, we build the training sample set by local similarity blocks, using PCA to extract the main features of the signal. In the domain of PCA, a threshold principle is given based on linear minimum mean square error estimates. Noise is suppressed by the threshold. Finally, we analyze the bias caused by the logarithmic transformation. The denoised image is obtained by bias estimation and the exponential transform. Experiment results indicate that the proposed method can effectively remove multiplicative noise while well preserve details of the image features.