Adaptive Importance Sampling Unscented Kalman Filter With Kernel Regression for SAR Image Super-Resolution
Sithara Kanakaraj, Prof. (Dr.) Madhu S. Nair, Saidalavi Kalady · IEEE Geoscience and Remote Sensing Letters · 2020
Resolution enhancement of Earth’s images from synthetic aperture radars (SARs), used for applications that require scene interpretations and detailed analysis, fails due to the presence of inherent speckle noise. An inexpensive alternative solution to the problem is to use super-resolution (SR) algorithms that deal with speckle. A novel approach to augment kernel regression into the Adaptive Importance Sampling Unscented Kalman Filter (AISUKF) framework for SAR image SR has been presented in this letter. We have experimented with three different nonlinear kernel regressions, namely, arc-cosine kernel, radial basis function kernel, and steering kernel (SK) regressions. Empirical results suggest that AISUKF with SK regression is more appropriate for the abovementioned SR problem resulting in a better denoised and more detail-preserved output.