Unfortunately because seismic data is band-limited, cross-correlelograms have multiple local maxima, and in noisy data situations, the maxima may have similar amplitude. Simply picking maxima is, therefore, prone to cycle-skipping problems. The problem of cycle-skipping is fundamental to the process of automated (and even manual) seismic interpretation. Rickett and Lumley (1998) addressed it by heavy smoothing and median filtering of the warp functions before the image resampling itself.
In this paper, I exploit the link between shaping filters and cross-correlation functions to incorporate the smoothing within the matching process. I pick shaping filter maxima as opposed to cross-correlelogram maxima, reducing the need for ad hoc smoothing after the fact.