Integrating knowledge from computer vision, coding and human-computer interaction to achieve ultra-compression of video data

Impact summary

Video has the highest capability to engage people than any other media type with the brain processing video 60,000 times faster than text. Research predicts that 80% of global Internet consumption will be video content by 2019 and 75% of mobile traffic will be video by 2020.

The increased need for video content stems from technologies such as augmented/virtual/mixed reality, CCTV cameras, mobile devices, wireless communication, Internet of Things (IoT), eye/face tracking technology and sensing brain signals. The challenge faced is how to manage the huge volume of video data with ultra-compression required for real time transmission and processing. To enable this, Charles Sturt University researchers integrated knowledge from computer vision, coding and human-computer interaction to achieve ultra-compression of video data.

The research team used pattern-based techniques to approximate arbitrary object shape and effectively track block level motion without explicitly segmenting foreground and background. The techniques are recognised as ‘asymmetric block partitioning’ in the international video coding standards. For example these new methods are used in HEVC (High Efficiency Video Coding), China-based AVS (Audio and Video Coding Standard) and Google-based VP9 standards.

The research team also developed background frame modelling techniques that capture dynamic scene background in real time. This technique can be used in:

(i) extracting common information from hyper-spectral images
(ii) modelling background frame for video coding
(iii) hole filling in virtual view synthesis for 3D scene generation
(iv) anchor frame to control error propagation for transmission

This concept is recognised by partially adopted as a ‘golden frame’ in the Google-based VP9 standard.

Impact date	2013
Category of impact	Economic Impact, Quality of life Impact
Impact level	Adoption