Polarimetric synthetic aperture radar speckle filtering by multiscale edge detection

Abstract

Polarimetric synthetic aperture radar (PolSAR) images are disturbed by an inherent noise having multiplicative properties called “speckle.” This noise is undesirable, and its treatment is difficult. To reduce the speckle, a polarimetric filtering is necessary to improve the image quality. The purpose of PolSAR filtering is to use the polarimetric information in the different channels to develop an efficient algorithm adapted to this data type, to reduce well the speckle and preserve the contained information. We present the PolSAR wavelet filtering applying the stationary wavelet transform: filtering by multiscale edge detection with two improvement techniques of wavelet coefficients, filtering by wavelet thresholding using the hard and soft thresholding and their two enhanced versions. Our contribution is based on the adaptation of wavelet thresholding to PolSAR data and on improvement techniques to filter polarimetric covariance or coherency matrix elements and span. The methods are applied to the fully polarimetric RADARSAT-2 images acquired over Algiers, Algeria (C-band), to the three polarimetric E-SAR images acquired on Oberpfaffenhofen area located in Munich, Germany (P-band), and to the simulated PolSAR images (L-band). We evaluate the performance of each filter based on the following criteria: smoothing homogeneous areas, preserving contours, and polarimetric information. Experimental results and a comparative study are included.