SAR image compression using optimum threshold based support vector machines

Synthetic Aperture Radar (SAR) systems used on an unmanned aerial vehicle (UAV), aircraft or spacecraft are used to build SAR images using the concepts of backscattering. SAR images consist of high-resolution reflected returns of radar frequency energy from terrain illuminated by directed beam of pulses generated by the SAR system. In the case of aircrafts, these SAR images are displayed on the cockpit, transmitted to ground station as well as stored in on-board storage disks. SAR images serve as vital source of information for a large variety of applications which includes reconnaissance, automatic target recognition and mapping of geographical area. There is an urge to compress SAR images which would in turn reduce the transmission time of SAR images from on-board systems to ground station. SAR image compression will also reduce the memory requirements on-board. In this paper, two approaches for SAR image compression using optimum threshold based Support Vector Machine (SVM) regression are proposed. The first approach uses only proposed SVM regression, whereas the second approach is a two-stage SAR image compression using standard 1-level wavelet decomposition followed by proposed SVM regression. In order to assess the efficacy of the proposed system, datasets from USC-SIPI image databases are employed and 67.81-77.36% image compression is achieved with a PSNR ranging between 36.20dB and 43.12dB.