MediaMatrix: A Video Stream Retrieval System with Mechanisms for Mining Contexts of Query Examples

Abstract

This paper presents a context-based whole video retrieval system that retrieves an entire video stream, not just individual scenes in videos. The main feature of this system is a novel query processing paradigm that involves data mining techniques to extract a query context, which is expressed as a combination of multiple videos. A query for this system consists of an example video and a context video. The example video is for representing the color-emotions hidden within video content, which are extracted using methods developed for color psychology research. The context video is for performing a context relevance computation between the example video and videos in a database. This paper demonstrates an implemented system that includes a SQL-like video stream query language and mechanisms for visualizing emotive contexts of a video stream.