Salient stills

Abstract

The transformation of a sequence of moving images to a still image presents several problems: a loss of perceived image quality when looking at a single video frame; and a loss of context, since no individual frame captures everything conveyed in a sequence of images. We address these problems by developing a structured representation of moving image data. Using this representation we manipulate the video signal to build salient still frames, including some images with enhanced resolution. We discuss methods of automatically building a structured representation, methods for building salient stills and the potential applications of these images.