A Video Indexing Ontology Using Fuzzy Metadata

点击翻译按钮获取中文摘要

A Video Indexing Ontology Using Fuzzy MetadataVideo has been very difficult to understand by a machine. While humanshave a seemingly "direct" way of seeing something and understanding ascene in terms of background, foreground objects, and motions, videounderstanding has been one of the perplexing problems of automatic videoand image analysis to date.This project proposes a "divide and conquer" approach, where severalhundred general-purpose concepts (e.g., outdoors, animals) will be usedto describe and annotate a very large universe of scenes commonlydepicted in video. Analogously to a limited vocabulary set of indexingterms one might find in a library card catalog, each video scene can beannotated through a combination of these concepts ("metadata"). To gobeyond a mere listing of objects, actions and scenes visible in thevideo, carefully chosen concepts also allow the description ofrelationships between them ("ontology"), which allows for much richercomposite descriptions. The challenge will be to define these conceptsso that they satisfy several criteria at once:* The concepts must represent things frequently visible in videobroadcasts.* The concepts must be clearly identifiable to give computeralgorithms a chance to detect them automatically.* The concepts must be linkable into an ontology that defines howconcepts are related.Since video annotations, whether done by a computer or a human, alwayswill contain errors, this work will incorporate probabilistic confidencemetrics ("fuzzy metadata") into the annotation. No existing indexing andclassification schemes have explicitly defined standards for measuringand reporting errors and omissions of indexing annotations; sincelibrarians and archivists have traditionally assumed that an indexcontains only complete, trusted and verified metadata.Furthermore, the project will assess to what extent these concepts canbe automatically extracted with state of the art video analysistechniques. Using footage from documentaries and television news, theproject will perform video search and retrieval experiments to determinethe usefulness of the ontology and the confidence of the annotations.URL: http://www.informedia.cs.cmu.edu/ontology