VideoWeb: Space-time aware presentation of a videoclip collection

Browsing to identify frame contents in a large collection of videoclips is tedious and time consuming if annotation or indexes are lacking. This paper presents a novel approach for presenting relationships among videoclips in a space-time aware manner, using a structure we call a VideoWeb. To construct such a VideoWeb, each individual clip is parsed into structural elements in accordance with its semantic meaning. Spatio-temporal correlations are then inferred among clips, based on structural similarity, in a whole-to-part manner. These correlations result in a set of strong relationships linking temporal sequence and spatial views, and weak relationships of spatial co-occurrence; both are graphically represented on a navigation board. Visualizing connectivity among clips provides clues enabling users to easily browse the videoclip collection in a purposeful manner. A VideoWeb faithfully represents the relationships among the videoclips, and it can be interactively manipulated to explore clips in a collection. Experimental results and user studies show that our approach can greatly facilitate browsing of a videoclip collection.