University of California, Santa Cruz professor James Davis says small-scale image-analysis technology is not yet refined enough to produce accurate and reliable results, despite advances in machine vision over the past several decades.
Davis receives requests from entrepreneurs seeking to resolve problems with image analysis, to enable them to perform functions such as comparing cell sample photos to stored images of different pathological cells. These entrepreneurs typically find that computer vision is roughly 20 years behind the level of advancement they expect.
"It just doesn't work as well as they need it to in order to get reliably accurate results," Davis says.
The best current approach is to use computer algorithms for preliminary searches and then use human employees to verify results. Davis is using a CITRIS seed grant to improve the interface between the automated algorithmic component of visual searches and the human part, using engineers to write code that in some places calls for human verification.
In addition, Davis' colleague on the project, University of California, Merced professor Ming-Hsuan Yang, intends to use Davis' work to hasten his own research. Yang and his team are creating super-resolution machine-vision systems that can zoom in on and clarify images.
From CITRIS Newsletter
View Full Article
Abstracts Copyright © 2013 Information Inc., Bethesda, Maryland, USA