Title: Powerpoint slides of Prof. Toby Berger's Shannon Lecture at ISIT 2002
Author: Toby Berger

Details: The slides from the lecture are divided into three parts and can be downloaded from the links below.  Slides with the proof of the main theorem that was not presented in the lecture is also available for download here.

Part 1: Introduction (2.0 MB)
Part 2: Neurons (5.8 MB)
Part 3: Neural Coalition and Main Theorem (3.5 MB)
Appendix: Proof of Main Theorem (0.4 MB)

 Back to Summary

Title: QVIX/CU30 Video Conferencing Transport Layer Development
Project Authors: Alan Choi and Edmund Fung

Abstract: In the Qvix/CU30 video conferencing application, the data stream plays an important role in determining the transmission rates, quality and privacy of the frames.  Therefore, this project's intent is to design and implement algorithms to enhance the efficiency and security of the network data stream.
 
Because one of the main objectives of Qvix/CU30 application is to maintain a 30 frames per second rate, congestion control is a necessity.  Without an effective congestion control, frame rate could drop below 25 fps and result in undesirable image quality.  In order to overcome this problem, a reliable TCP communication channel was employed for sending feedback control messages to the sender regarding the occurrence of lost frames. Subsequently, the frame rate is adjusted to appropriate settings.

In order to provide a secure transmitted data stream, frame encryption was used. However, since another main objective of Qvix/CU30 application is to achieve low latency in frame display, the encryption method must be chosen to meet a demanding latency requirement operating at 30 fps.  It was found through testing that these requirements could be satisfied by Advanced Encryption Standard (AES).

This report will discuss the design and implementation of the transport layer development. Experiments were conducted and the test results are included as supportive evidences of the features.  Source codes are also attached for reference.

 Back to Summary

Title: A Software-Only Videocodec using Pixelwise Conditional Differential Replenishment and Perceptual Enhancements
Author: Yi-Jen Chiu and Toby Berger
Published in: IEEE Transactions on Circuits and Systems for Video Technology, April 1999

Abstract: Designing a video codec involves a four-way tradeoff among computational complexity, data rate, picture quality, and latency.  Rapid advancement in VLSI technology has provided CPU's with enough power to accommodate a software-only video codec.  Accordingly, computeational complexity has resurfaced as a major element in this tradeoff.  With a view toward significantly reducing computational complexity relative to standards-based video codecs, we introduce a pixelwise conditional differential replensihment scheme to compress video via perception-senstive decomposition of difference frames into a facsimile map and an intensity vector.  Our schemes, which apply techniques from facsimile, as transform-free.  Some of them also involve no motion compensation and hence are completely free of block-based artifacts and particularly computationally economical.  The fusion of our facsimile-based video coding schemes and spatio-temporal perceptual coding techniques facilitates powerful software-only video conferencing on today's medium and high-end personal computers.  Indeed, assumng a frame capture driver has been provided, out motion-compensation-free approach as yielded a software-only, full duplex, full color video conferencing system that conveys high quality, CIF/QNTSC-sized video at 30 frames per second on 200 MHz Pentium PC's sending less than 300 Kbps in each direction.

We also present new spatio-temporal compression techniques for perceptual coding of video.  These techniques, motivated by the classical psychological experiments which led to formulation of the Weber-Fechner law, allow video codec systems to capitalize on peoperties of the human visual system.  Some of our spatio-temporal perceptual techniques apply not only to our proprietary pixelwise conditional differential replenishment schemes we describe for video conferencing but also can readily be incorporaed into today's popular video standards.

 Back to Summary

Title: Multiple Descriptions Encoding of Images 
Author: P. Subrahmanya and Toby Berger 
Published in: DCT

Abstract: Consider the following multiterminal data compression problem. An image is to be compressed and transmitted to a destination over a set of unreliable links. If only one of the links is functional (i.e. data transmitted over all other links is completely lost), we would like to be able to reconstruct a low resolution version of our original image. If more links are functional, we would like the image quality to improve. Finally, with all links functioning, we desire a high resolution, possibly lossless reconstruction of the original image. This problem is related to the multiple descriptions problem in source coding literature. See [1] for a survey of the literature.

Several motivating applications exist for this problem. For instance, in a digital image archive, it is desirable to store low resolution versions of each image at multiple locations so that even if a few of them are unavailable/corrupted/destroyed, we can reconstruct approximations to the original image using those that survive. If all the low resolution versions are available, we would like to be able to reconstruct the original image losslessly.

Another example arises in the downloading of images from an image library over the internet. Traditionally, all images are stored at a single site in the network. If any part of the network route leading to a site is congested, viewers wishing to view an image on that site will be faced with debilitating delays. A solution to this problem is to place low resolution versions of the image at multiple sites. Even if some of the sites are congested, viewers can satisfy themselves with a lower resolution version of the image reconstructed using the sites that become available within a reasonable delay. If all the sites can be accessed, then the viewer would get a high resolution version of the image. Usually, lossy compression is used in this application in order to achieve the high compression ratios needed to use network bandwidth effectively.

We present conceptually simple and computationally efficient schemes that are useful in scenarios such as those described above. Our schemes can be implemented as simple pre-processing and post-processing operations on existing image compression algorithms.

Reference: [1] V.Vaishampayan, Design of multiple description scalar quantizers, IEEE Transactions on Information Theory, IT-39(3):821-834, May 1993.

 Back to Summary

Project Title: Computer-based Video Telephone Over 28.8 kb Modem Link  
Author: Jay Chunsup Yun

Abstract: Over the past few years, several M.Eng projects concentrated on developing techniques for a video telephone system that runs on a computer and communicates over a regular telephone line. These project yielded encoder and decoder software with some performance limitations. While the decoder is able to provide real-time decoding at 10 to 15 frames per second, the encoder runs too slow for a realistic application. This continuation project focuses on enhancing the execution time of the encoder software and on preparing for full-duplex operation.

The project has three main objectives:

1. Improving the encoder algorithms for faster and simpler executions.
2. Combining the encoder and the decoder into a single application.
3. As time allows, designing techniques for problems not yet dealt with.

The frame rate of the resulting encoder algorithm is increased by over three times. Complete, full-duplex operation is not achieved in this project, but the video telephone software is able to encode and decode concurrently. The project leaves many detailed suggestions for future development.

 Back to Summary

Project Title: Object Tracking with Dynamic User Override
Author: Worawat Tangsangasaksri

Abstract: A program was developed to demonstrate the performance of object tracking in the motion compensated video codec. The video compression program is based on the software modules developed by members of the DISCOVER lab; the modules werecombined with a display function to show the result of the decoded image frame by frame as in real-time processing. A mouse-driven user interface was developed to provide a convenient means for initializing the object tracker box andenabling the user to reset the box location when the object tracking algorithmfails. An enhanced algorithm for motion estimation that improves the processing speed with negligible quality degradation is introduced.

 Back to Summary

Project Title: A Vector Quantized Design for Video Sequences Using a Deterministic Annealing Approach to Codebook Generation
Author: Wayne G. Phoel

Abstract: This project employs deterministic annealing to determine the codebook for encoding difference frames of head images isolated from a video sequence. The purpose of this project is to attempt to encode image sequences in a manner that will meet the existing need for high quality inter-office video conversation.

 Back to Summary

Project Title: Occlusion Detection and Edge Processing in a Correlation Based Object Tracker 
Author: Neilesh R. Patel 

Abstract: A robust object tracker has been developed for use in video conferencing codecs. This tracker operates on the principle of hierarchical block matching. Objects are able to be tracked as they translate, moderately rotate, or slightly deform.

 Back to Summary

Project Title: Workstation/Modem Interfaces for Asynchronous Video Data Transfer
Author: Elya J.Kapelyan

Abstract: This project deals with issues related to the design and implementation of terminal software for high-speed (28.8 Kbps) transfer of compressed video frames over telephone line. In particular, methods of parallelizing time-critical routines within the UNIX operating system are discussed.

 Back to Summary

Project Title: Reliable On-Line Human Signature Verification Systems for Point-of-Sale Applications
Author: T. Berger, E. Aviczer and L. L. Lee
Published in: IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 18, No. 6, June 1996, 643-647

 Back to Summary