Recover Password  New User
Undergraduate Research Project Management System

Speeding up Image Compression Research via UAA's GPU Server and Evolutionary Computation

Status Current
Seeking Researchers No
Start Date 08/01/2011
End Date 05/30/2012
Funding Source Undergraduate Research Grant
Funding Amount
Community Partner
Related Course
Last Updated 09/21/2011 11:12PM
Keywords evolutionary computation, image compression, GPU

People

Faculty
  Frank Moore, Kenrick Mock

Student Researchers
  Brendan Babb

Abstract

Image compression is used on a day to day basis by nearly everyone. The JPEG 2000 standard is used to compress images that appear on websites, are stored on digital cameras, are displayed on cell phones, are transmitted by satellites, etc. The JPEG 2000 standard uses a wavelet for compressing and decompressing images. Previous research at UAA has been successful in evolving wavelet-like transforms that improve image compression. Recently UAA was awarded NASA funding to look into developing better image transforms and hyper-spectral transforms. NASA funds have been used to purchase Academic Matlab licenses and Toolboxes to perform research runs on the UAA computer. Prior research used standalone computers for small, preliminary tests, but required multiple processors on the ARSC supercomputer at UAF in order to complete large scale tests. Evolutionary computation performs similar calculations across many processors; the more processors you have the faster it works. Graphical Processor Units (GPUs) also have many processors within a single card. Due to widespread use by the multi-billion gaming industry, GPUs have rapidly developed to the point where they are now capable of outperforming supercomputers for numerical calculations, by factors in excess of 100:1. It is now possible to buy a relatively cheap graphics card and insert it into an existing server to create a very fast supercomputer for performing evolutionary computation tasks in significantly less time. The goal of this project is to use GPUs and evolutionary computation to evolve significantly better wavelets for image compression and to serve as a model for additional UAA research that can leverage massive parallel computing.

Shared Project Files (e.g. papers, presentations)

File name Description Uploaded by