报告人简介
Dr. Zhi Ding (S'88-M'90-SM'95-F'03, IEEE) is a Professor of Electrical and
Computer Engineering at the University of California, Davis. He received his
Ph.D. degree in Electrical Engineering from Cornell University in 1990. From
1990 to 2000, he was a faculty member of Auburn University and later,
University of Iowa. Prof. Ding has held visiting positions in Australian
National University, Hong Kong University of Science and Technology, NASA Lewis
Research Center and USAF Wright Laboratory. His major research interests lie in
the general field of signal processing and communications. Prof. Ding has
active collaboration with researchers from many universities including those in
Australia, China, Finland, Japan, Canada, Taiwan, Korea, and Singapore. He has
coauthored over 400 technical papers and two books. Dr. Ding is a coauthor of
the text: Modern Digital and Analog Communication Systems, 5th edition, Oxford
University Press, 2019.
Dr. Ding is a Fellow of IEEE and has served on technical programs of
a number of workshops and conferences. He served both as a Member and also the
Chair of the IEEE Transactions on Wireless Communications Steering Committee
from 2007-2001. Dr. Ding was the Technical Program Chair of the 2006 IEEE
Globecom and the General Chair of the 2016 IEEE International Conference on
Acoustics, Speech and Signal Processing (ICASSP). He served as an IEEE
Distinguished Lecturer (Circuits and Systems Society, 2004-06, Communications
Society, 2008-09). He received the 2012 Wireless Communications Recognition
Award from the IEEE Communications Society. He currently also serves as
the Chief Information Officer of the IEEE Communications Society.
报告摘要
JPEG2000 (j2k) is a highly popular format for image and video
compression. With the rapidly growing applications of cloud based image
classification, most existing j2k-compatible schemes would stream compressed
color images from the source before reconstruction at the processing center as
inputs to deep CNNs. We propose to remove the computationally costly
reconstruction step by training a deep CNN image classifier using the CDF 9/7
Discrete Wavelet Transformed (DWT) coefficients directly extracted from
j2k-compressed images. We demonstrate additional computation savings by
utilizing shallower CNN to achieve classification of good accuracy in the DWT
domain. Furthermore, we show that traditional augmentation transforms such as
flipping/shifting are ineffective in the DWT domain and present different
augmentation transformations to achieve more accurate classification without
any additional cost. This way, faster and more accurate classification is
possible for j2k encoded images without image reconstruction. Through
experiments on CIFAR-10 and Tiny ImageNet data sets, we show that the
performance of the proposed solution is consistent for image transmission over
limited channel bandwidth.