Decomposing GPR images in a variation approach

Abstract

Interpreting a ground penetration radar (GPR) image is an important task in research into subsurface conditions. In a GPR image, noise usually obscures the weak reflections, especially for objects buried at deep locations. The discrete wavelet transform has been recognized as an efficient tool for depressing noise effects appearing in a GPR image. Another approach is the total variation (TV) de-noising model, which was first introduced by Rudin, Osher, and Fatemi in 1992. This paper uses the TV de-noising model to remove noise from GPR images. By computing the signal-to-noise ratios, the performances of the discrete wavelet de-noising and TV de-noising models are evaluated. From the experimental results, the TV de-noising model offers an efficient, numerically stable, and robust way to deal with noise present in a GPR image.