next up previous [pdf]

Next: Acknowledgments Up: Liu etc.: Structurally nonlinear Previous: Field Data Examples


We have introduced a new technique that combines structure prediction and signal-enhancing filters: similarity-mean filter and lower-upper-middle (LUM) filter, for attenuating random noise while enhancing structural information. The technique is generic. One can replace the proposed filters with other prediction and signal-enhancing filter methods. Experiments with synthetic and field data show that nonlinear structure-enhancing filters can both eliminate random noise and preserve fault information in seismic images. Our method makes an implicit assumption that data is made of long, coherent events and random noise. It may need a further improvement to deal with areas where either crossing dips or no coherent events are present.