Polarimetric SAR Image Filtering by Infinite Number of Looks Prediction Technique

Abstract

Speckle filtering in synthetic aperture radar (SAR) and polarimetric SAR (PolSAR) images is indispensable before the extraction of the useful information. The minimum mean square error estimate of the filtered pixels conducted to the definition of a linear rule between the values of the filtered pixels and their variances. Hence, the filtered pixel for infinite number of looks (INL) is predicted by a linear regression of means and variances for various window sizes. In this article, the infinite number of looks prediction (INLP) filter is explored in details to emphasize its ability to reduce speckle and preserve the spatial details. Then, the linear regression rule has been adapted to PolSAR context in order to preserve the polarimetric information. The number of the processed pixels used in the linear regression is adjusted to the variability of the scene. This effort increased the filtering performances. The reduction of the correlation between the pixels which constitutes an additional filtering criterion is discussed. Compared to the initially applied filter, the results showed that the improved INLP filter increased in speckle reduction level, augmented the preservation of the spatial details, increased the spatial resolution, reduced the correlation between the pixels and better preserved the polarimetric information. Simulated, one-look and multilook real PolSAR data were used for validation.