Exploiting Human Color Discrimination for Memory- and Energy-Efficient Image Encoding in Virtual Reality

A VR image compression algorithm reducing DRAM traffic by 66.9% with minimal perceptual quality loss

Institute Reference: 1-24027

Background

Virtual Reality (VR) is emerging as a ubiquitous computing platform, with applications spanning healthcare, education, communication, training, and design. Efficient computing substrates are critical to VR’s progress, with DRAM access energy significantly impacting system energy consumption. Traditional framebuffer compression methods, which are numerically lossless, are not optimized for human perceptual quality.

Technology Overview

This technology leverages human color discrimination abilities to create a perceptually lossless, but numerically lossy, image compression system for VR. The algorithm adjusts pixel colors to minimize bit encoding costs without introducing visible artifacts. It exploits the fact that humans cannot distinguish colors that are close to each other, especially in peripheral vision. Implemented as lightweight hardware extensions, this system significantly outperforms existing framebuffer compression mechanisms.

Further Details:

Benefits

  • Energy Efficiency: Reduces DRAM traffic by 66.9%, leading to substantial energy savings.
  • Performance: Outperforms current framebuffer compression methods by up to 20.4%.
  • Quality: Maintains high perceptual quality with minimal visible artifacts, validated by psychophysical studies.

Applications

  • Virtual Reality Systems: Enhances the performance and energy efficiency of VR headsets and other devices.
  • Mobile Devices: Applicable to any mobile system requiring efficient image compression.
  • Augmented Reality (AR): Can be adapted for AR systems to improve user experience and device longevity.
URV Reference Number: 1-24027
Patent Information:
For Information, Contact:
Curtis Broadbent
Licensing Manager
University of Rochester
585.273.3250
curtis.broadbent@rochester.edu
Inventors:
Yuhao Zhu
Nisarg Ujjainkar
Ethan Shahan
Kenneth Chen
Qi Sun
Budmonde Duinkharjav
Keywords: