SAR image despeckling and compression using K-nearest neighbour based lee filter and wavelet

Abstract

Synthetic Aperture Radar (SAR) is an important tool for obtaining information from the surface of the Earth with the capability of producing high resolution images. SAR has the advantage of all-weather sensing and it has tremendous applications in areas such as the military, agriculture, oceanography, resource mapping, and so on. However, SAR image is contaminated by speckle which is multiplicative in nature, and its presence in the image makes image contents interpretation difficult. Also as tremendous amounts of data gathered from the SAR system increases, the storage capacity and transmission speed do not increase at the same rate. In this paper we present a SAR image despeckling and compression technique. In this technique, the K-Nearest Neighbour (KNN) algorithm is employed to modify the well-known Lee filter to improve its performance for SAR image despeckling. The despeckling method is used to address some of the shortcomings of existing SAR image despeckling filters. Existing speckle filters introduce blur when used to despeckle SAR images. In addition, feature and edge preservation problems also arise. With our method, a suitable number of nearest neighbour pixels within the sliding window can be selected for calculating the filtering parameters. With a 5×5 window size, the best result is obtained when only 15 out of the 25 pixels are used. The improved filter is then used to despeckle the SAR image before it is compressed using the Two-Dimensional Discrete Wavelet Transform (2-D DWT). The wavelet transform coefficients are coded with the Set Partitioning in Hierarchical Trees (SPIHT) scheme with the removal of the arithmetic coding stage. It is evident, from the simulation results shown that better results are obtained when a SAR image is despeckled prior to compression.