A Bivariate Thin-Plate Adaptive Smoothing Spline (BTPASS) to reduce noise in photogrammetric digital surface models

Photogrammetric surface restitution has considerably improved, with advances in stereo-matching algorithms allowing dense and relatively clean digital surface models to be created from triangulated stereo-imagery. Nevertheless, application to scanned historical analogue imagery in

challenging contrast is exposed to noise due to film grain. Restitution can produce ancillary information about the intersection error from homologous rays that measure the quality of the match. This paper proposes and demonstrates a post-processing method to filter the resulting surface with thin-plate smoothing spline. The level of smoothing is locally adaptive to allow the topography to be preserved when correctly restituted, while significantly reducing noise on

low-contrast areas.