Steering kernel regression: An adaptive denoising tool to process GPR data

Abstract

The recently introduced steering kernel regression (SKR) framework was originally developed to attenuate random noise in images and video sequences. The core of the method is the steering kernel function which incorporates a robust local estimate of image structure into the denoising framework. This helps to minimize image blurring and to preserve edges and corners. As such filter characteristics are also desirable for random noise attenuation in ground-penetrating radar (GPR) data, we propose to adopt the SKR method for processing GPR data. We test and evaluate this denoising method using different GPR data examples. Our results show that SKR is successful in removing random noise present in our data sets. Concurrently, it preserves local image structure and amplitudes. Thus, the method can be considered as a promising and novel approach for denoising GPR data.