Efficient Video Filtering of MPEG-7 Streams

The production of digital multimedia content is continuously increasing. With the advent of the digital cable, satellite and terrestrial televisions, thousands of channels are practically available for users. Moreover, multimedia is widely used in many professional applications, such as video program production, video surveillance, and e-learning. Consequently, the use of techniques for video retrieval as well as video filtering is becoming of crucial importance. The adoption of the MPEG-7 standard is a significant step forward in simplifying the video retrieval and filtering. However, the performance issue can be relevant if the retrieval must be accomplished in real time, as in some applications such as the video surveillance or video filtering in general. Moreover, the number of streams to search, the number of queries, and the computational complexity of the feature similarity measure can heavily affect the effectiveness of such real time filtering applications. In this paper we present the Pivoted Stream, a novel approach for efficient filtering of a video stream by using the MPEG-7 descriptors. Our proposal exploits the properties of the metric spaces, in order to reduce the computational load of the filtering receiver.