A METHOD AND APPARATUS FOR MULTIMODAL STORY SEGMENTATION FOR LINKING MULTIMEDIA CONTENT

摘要

Stories are detected in multimedia data composed of concurrent s treams for different modes such as audio, video and text and linked to related stories. First, time periods of uniformity in attributes of the streams serve as "building blocks" that are consolidated according to rules that are characteristic of the story to be detected. The attributes are then ranked by their respective reliabilities for detecting the story to be detected. An inter-attribute union of the time periods is cumulated attribute-to-attribute in an order that is based on the ranking. A buffered portion of the multimedia data that is delimited by the starting and ending times is retained in mass storage. The starting and ending times are indexed by characteristics of content of the portion to form a story segment which is maintained in a data structure with links to related story segments.