Improvement of PWF filter using wavelet thresholding for polarimetric SAR imagery

Abstract

The images acquired by polarimetric SAR radar systems are characterized by the presence of a noise named speckle. This noise, have a multiplicative nature, corrompt at the same time the amplitude and the phase which complicates the data interpretation, degrades the performance of segmentation and reduces the targets detectability. From where need to pretreate images by adapted filtering methods, before carrying out their analysis. In this article, we study the polarimetric wightening filter PWF of Novak and Burl which treats the polarimetric covariance matrix to produce a filtered intensity image. We propose two methods to improve the PWF filter: the first integrates the technique of Lee edge detection to improve the filter performance and detect fine details of the image. This method is called LSDPWF (Lee Structure Detection PWF). After detecting the edges, we filter the detected regions in the polarimetric channels by the PWF filter. The second combines the method of filtering by wavelet thresholding with PWF filter using the stationary wavelet transform SWT. This method is called EPWF (Enhanced PWF). In the wavelet thresholding, we use the soft thresholding which sets to zero the amplitudes of coefficients that are below a certain threshold. So we propose to extend the wavelet thresholding, to apply it in polarimetric SAR images and use the polarimetric information to calculate the threshold on the wavelet coefficients. We implemented these filters and applied them to RADARSAT-2 polarimetric images taken on the areas of Algiers (Algeria). A visual and statistical evaluation and a comparative study are performed. The performance evaluation of each filter is based on smoothing homogeneous areas and preserving edges.