The goal of the MobileASL (American Sign Language) research project is to enable sign language communication over the US cellular network. State of the art video compression is achieved by using reference frames to predict future frames, taking advantage of the image correlation of neighboring frames. Unfortunately the cellular network in America is low bandwidth and lossy. Loss of data can greatly impact the quality of compressed video because of errors are propagated through predictions. I investigate techniques to minimize the effect of error loss on the intelligibility of transmitted ASL video. As both computation power and bandwidth are limited on cellular devices, I concentrate on determining good allocations of these resources. Specifically I focus on using feedback to prevent error propagation.
Advised by Richard Ladner
CSE 203
Wednesday
April 16, 2008
3:30 - 4:20 pm