Many seismic images, especially those acquired on land, are contaminated with random noise that impedes interpretation and interferes with further processing and analysis. This random noise can be recognized by its dissimilarity from trace to trace. Signal, on the other hand, is recognized by its lateral continuity within the image. In most cases, this continuity results from the sedimentary character of the data being considered.
The methods we consider here predict only linear events within a seismic image. While the human eye recognizes the continuity of nonlinear events, the mathematical tools available to us work best on linear problems. Even though many continuous seismic events are not linear, windowing the image into smaller areas makes most of these events at least approximately linear.
The two lateral prediction methods discussed here are referred to as f-x prediction and t-x prediction. Prediction in the f-x domain, as presented by Canales 1984 and expanded by Gulunay 1986, predicts linear events in the frequency-space domain and has become a standard processing tool. As an improvement over the f-x prediction technique, we propose a t-x prediction technique that predicts linear events in the time-space domain.