Data-driven Weighted Median Filter and Its Application to Structure-enhancing Filtering

Random noise attenuation is a persistent problem in seismic interpretation, most edge detection and coherence algorithms are often sensitive to random noise. We present a new structure-enhancing method, which aims at reducing random noise while protecting structural information. We use local dip pattern to form a structural prediction of seismic traces from neighboring traces. The local similarity between prediction traces and original traces can be calculated to design a data-driven weighted median filter. Numerical test using 2-D field data and comparison with structurally similarity-mean filter confirm the effectiveness of the proposed structure-enhancing filter.