Ontology for Linked Video Data

Traditionally, video is viewed and stored linearly on a frame-by-frame basis. This temporal linear restraint makes searching and cataloging video a difficult and time-consuming operation.

Utilizing RDFa metadata we can create a video ontology to automatically produce an accessible index. Using core primitives and relational indices over video content can be used across diverse domains. Starting with the tree structure of RDF and adding Graph Theory methods can be used where appropriate and we can exploit links and logic.

Our approach is to create linked video data by developing a multi-layer hierarchy that includes semantic references linking each of seven layers. At the lowest or most basic layer, layer 1, we designate, or tag, objects (such as a car, house, mountain, sky, person, dog, ect.) within a frame. The next layer, is a 1 second segment consisting of 25 frames and creates an Identity frame, or IFrame. Layer 3 is the shot or take frame. Layer 4 identifies locations. Layer 5 is for plot characteristics, and the layer 6 is the scene characteristics. The final layer 7 is for overall film identify and reference including items such as: the title, producer, director and cast designation.

Level 7 is closest to the existing Dublin Core Video such as:


This hierarchy of layers is interlinked at each level through metadata. As a result, the digital video can relate a wide variety of references in preparation for relevant advertisement placement material.

The Semantic concepts of this project can be used for both manual and automatic indexing of video footage. The representation scheme conforms to metadata standard MPEG-7 and contains only trusted and verified metadata. It goes beyond the Dublin Core and is extensible for specific applications. Since there are 90,000 frame per hour, adding tags to every frame requires large investment in effort. However, since so much information is redundant frame to frame that reduction techniques just like in compression software can yield significant improvements. The main method is to scan for changes in tags from frame to frame.

Thinking on the Web: Berners-Lee, Turing and Godel

Developing Semantic Web Services

Connections: Patterns of Discovery