Highly Efficient Scale-Space Based Hierarchical Video Indexing and Retrieval

EPSRC

Description:

The main objective of this project is the development of an efficient system for accessing video on the Internet and general video databases. Specifically, hierarchic video querying, video annotation, video cataloguing and interactive navigation over distributed networks and general video databases are the core research issues addressed in this project. This project builds on existing work, but extends it to achieve efficiency and robustness, which are the two major shortcomings of video search engines developed so far. The project aims to advance the state-of-art in terms of either proposing novel approaches or pushing forward the frontiers of current methods to meet the challenges of the system previously outlined. It will provide more efficient solutions to issues for which current approaches fail, including:

• Robust content-based video retrieval assuming the query video-object is roughly specified by the user
• Almost imperceptible time of response to the user query
• Content based query refinement and relevance feed-back at different levels of detail.

This project focuses on the efficiency and robustness of the envisaged system. These objectives will be achieved using a nonlinear scale-space paradigm to generate a family of successively simplified images keeping relevant scene information and building a hierarchically structured image-descriptor space. This is the major novel element of this project, that has not been proposed or developed before. In addition new scalable and highly efficient metrics to compare features in the descriptor space should be developed. Specifically, the final system will comprise:

• Extraction of key-frames representing long video sequences in different compressed domains including DCT, SA-DCT and wavelets
• Image-feature extraction in a family of suitably simplified images (scale-space evolution)
• Definition of hierarchically structured features and robust video-object descriptors
• Definition of scale-, translation- and rotation-invariant metrics in the descriptor space
• Adjustment of these descriptors to make them compliant with the forthcoming MPEG-7 standard
• Use of the feature hierarchy to provide highly efficient query-responses within large databases.
  

Funding

EPSRC  funded project, 2000-2002 £ 59,000

Scientist in charge

Ebroul Izquierdo

People Involved

Janko Calic