Video Data Compression

Video data compression is concerned with reducing the amount of data required to reproduce a digital video. It is a key component in facilitating the widespread use of digital video, which is currently prevented by the mismatch between the huge storage and transmission bandwidth requirements of video and the limited capacity of existing computer systems and communications networks. Although the range of video compression standards now available, such as H.261 or MPEG-2, exhibit acceptable performance when operating at high bit rates (eg >64 Kbps), they perform badly at very low bit rates, prompting the need for the development of the next generation of video compression standards.

We are investigating algorithms for video data compression at very low bit rates (<64 Kbps). Our work is primarily concerned with model-based techniques, in which a video sequence is represented in terms of objects or surfaces with associated colour, texture and motion parameters. This approach to video coding is a central component in the planned framework for the MPEG-4 compression standard and if successfully implemented would give rise to both very low bit rates and representations which are well suited to the envisaged multimedia applications of data compression, such as interactive video, generation of special effects, video editing, merging animation and video, etc.

We currently have two projects working in this area. The first employs a 2.5-D scene model consisting of surface patches and object boundary segments. The motions of the patches and boundaries are estimated using a combined region and feature-based approach, in which local 2-D motions are modelled in terms of a 6 parameter affine model and linear feature estimates are used to enhance the estimation in the vicinity of object boundaries. Both components are implemented within a multiresolution framework based on a generalised wavelet transform. More recently, a second project is looking at extending this approach to allow 3-D information in the form of surface orientation and 3-D motion to be utilised in order to improve the effectiveness of the coding strategy.

Staff and Students

Andrew Calway, Stefan Kruger.

Publications

Collaborators

Image and Signal Processing Group, University of Warwick.

Support

The research is supported by the EPSRC and NDS Ltd.