An effective approach to image/video compression and analytics

Project Title:

An effective approach to image/video compression and analytics

Supervisors:

  • Professor Manzur Murshed (Principal)
  • Associate Professor Shyh Wei Teng
  • Professor Guojun Lu
  • Dr Suryani Lim (ECR)

Contact person and email address:

A brief description of the project:

Images and video have two major characteristic: very large data size and lack obvious semantic structure. These two characteristics lead to two separate and distinct research tracks developed to deal with images and video. The first track of research is on image and video compression, aiming to compress images/video as much as possible without much consideration of how the images and video would be used eventually. Some examples of this track of research and standards are JPEG, MPEG and ongoing research efforts in image/video compression. The second track of research is on image/video analysis, indexing and retrieval, aiming to determine features and semantics automatically, and then search and find relevant images effectively. Some major research efforts in this area include image segmentation, feature representations, image annotation, similarity measurements, and deep learning methods.

As a result of the above two separate tracks of research, images/video must be decompressed first before image analytics can be applied, leading to low efficiency and infeasibility of real time implementation. Worse still, some information needed/useful for end applications may be permanently lost during the compression process.

This project aims to develop a method to integrate these two tracks of research such that image/video contents can be tracked and searched effectively and efficiently without decompression. Specifically, the project will develop an adaptive video coding technology generating metadata on-demand to improve efficiency and effectiveness of video analytics in compressed domain.