Video chatting and on-demand TV have gone from novelty to daily use in just a few short years. As people increasingly share photos and stream video, the challenge is finding an easier way to transmit those gigabytes over bandwidth-limited networks. For almost 20 years, IEEE Fellow Sheila Hemami has focused on this problem.
“We would all love to have infinite bandwidth, but we don’t,” says Hemami, a professor and chair of the electrical and computer engineering department at Northeastern University, in Boston. “Additionally, communication links are not 100 percent reliable, so service providers have to think about what to send and how to send it.”
The idea, of course, is to send images using the least possible bandwidth but without compromising their quality. The goal is to reproduce a digitally transmitted image that looks good at the other end. Hemami’s research consists of learning how best to do this when transmitting visual data over digital channels such as the Internet, cellular networks, and satellite links constrained in bandwidth and quality of service. Her work applies to images captured by cameras as well as medical imaging technology, such as MRI and CT scans. “I am primarily interested in material that includes video and images that people are going to watch for entertainment, rely on for diagnosing illnesses, or refer to in order to make a decision such as what house to buy,” she says.
This research interest extends to Hemami’s volunteer activities with IEEE. During her more than 25 years as a member, she has served on many program and conference-organizing committees covering signal and image processing, compression, and perception.
As the recently elected 2015 vice president of IEEE Publication Services and Products, she helps lead the strategy for producing, delivering, and ensuring the integrity of IEEE’s information services and products. She also serves on The Institute Editorial Advisory Board. Meet the rest of the board.
OUT OF SIGHT
Humans are imperfect sensors, Hemami says. “We have sloppy color vision. We’re much less sensitive to color errors than to structural ones.” Today’s image and video compressors take advantage of this to reduce memory usage and network bandwidth. Commonly used compression formats such as JPEG, for instance, do not save and reconstruct images bit by bit but strategically discard information such as soft transitions in color hue, which the average human eye cannot perceive.
By better modeling human visual perception and using more-sophisticated encoding strategies, Hemami has developed compression software that gets rid of even more data that people don’t see. People, for instance, tend to spot errors in image areas containing sharp edges and strong textures. So one way to retain image integrity while cutting bit rate is to keep fine details in those areas but make other areas more coarse.
Her software code, which can be incorporated into any standard JPEG compression engine, generates an image that appears the same to a person but is 30 percent smaller in byte size and thus uses 30 percent less memory. “In compression, even a 2 percent gain is considered very good, so 30 percent is tremendous,” she says.
Hemami has spent years trying to understand how people perceive images. She has conducted many experiments that involved showing volunteers natural images that were strategically distorted to see which details people could and could not see.
In 2009, Hemami who was then at Cornell and her colleagues used similar techniques to develop video-compression software that let deaf people use sign language via video chat over mobile devices. Until then, the deaf community had only been able to use text messaging; limited network bandwidth had prevented the transmission of relevant details in hand and face movements over video chat. The software worked, in part, by distinguishing between areas of the image, such as the signer’s hands and face, that had to be in high resolution and areas that could be reproduced in low resolution while still getting the hand signals across clearly.
THE EARLY STAGES
Hemami grew up with a fondness for “math, science, and open-ended problem solving.” Her father was an electrical engineer, and studying engineering in college was a natural choice for her, she says.
After earning her bachelor’s degree in 1990 in electrical engineering from the University of Michigan, in Ann Arbor, she went on to get her master’s degree and Ph.D. from Stanford in 1992 and 1994, respectively. Her Ph.D. thesis addressed the reconstruction of compressed digital images and videos from data packets sent over digital networks.
During her last year at Stanford she joined HP Laboratories in Palo Alto, Calif., as a member of the technical staff. Her research group was doing exploratory work on what was then an exciting, unheard-of technology: video on demand. The project was a rudimentary precursor to the streaming video technology of today, she points out, but its practicality was limited by the computing power, memory, and transmission bandwidth available at the time.
In 1995, Hemami joined the faculty at Cornell’s School of Electrical Engineering. She directed the Visual Communications Lab there until 2013, when she joined Northeastern University.
Hemami became an IEEE student member as an undergrad. Her first leadership role was as chair of the Image and Multidimensional Signal Processing Technical Committee, from 2006 to 2007.
During her one-year tenure as vice president of Publication Services and Products, she says her goal is to ensure “that the IEEE Xplore Digital Library should be the first destination that tech professionals think of for high-quality technical content in science and engineering.”