Hybrid video system supporting content-based retrieval

Abstract

The article suggests a Hybrid Video System (HVS) which supports the semantic retrieval of all types of users and the similarity retrieval of indefinitely formed and very large amounts of video data. The HVS divides a set of video data into video documents, sequences, scenes and objects to model the metadata and suggests a Three layered Hybrid Object-oriented Metadata Model (THOMM) which is composed of the raw data layer for physical video stream. The metadata layer supports feature based retrieval and similarity retrieval and the semantic layer reforms queries. Grounded on this model, we suggest a video query language which makes content-based query and similarity query possible and Video Query Processor (VQP) to process the query. In particular, the similarity query processed by performing the annotation-based retrieval using the concept layer and then the feature-based retrieval using the object-feature layer. Thus search space and time can be reduced. Also we present the formula for degree of similarity. The suggested system is implemented with using the Visual C++, ActiveX and ORACLE under Windows NT. © 1999 IEEE.