On predicting visual popping in dynamic scenes

Abstract

Popping is a major source of visual artifacts in dynamic scenes. To alleviate or avoid it, usually some temporal smoothing scheme is employed or levels of detail are chosen conservatively based on geometric deviation measures. In this paper, we consider the actual perceptibility of popping artifacts and its prediction. We first discuss several issues affecting popping perception, pointing out its complexity. Introducing some simplifying assumptions, we then present a practical perceptually-motivated predictor for popping. It makes heavy use of a spatio-velocity color vision model and aggregates the model output in a novel and useful way. We demonstrate the predictor's application to concrete examples, and discuss a conducted user study which indicates the validity of our approach.