Filtering of dual-polarization radar images based on discrete cosine transform

It is well known that filtering of multichannel images is, in general, more efficient compared to separate filtering of component data. This property can be exploited in processing dual-polarization radar images acquired by synthetic aperture radar (SAR) which are distorted by intensive speckle. Filtering of dual-polarization SAR images composed of HH and VV polarization components is studied. A novel filtering method based on 3D discrete cosine transform in blocks with preliminary normalization of data in blocks is proposed. Three quality metrics - MSE, MSE-L for locally active regions, and MSSIM - are analyzed to compare performance of the proposed method to component-wise DCT-based filtering and despeckling that employs logarithmic homomorphic transform. Advantages of the proposed method are shown quantitatively for simulated images and for real-life TerraSAR-X data.