Mari
Ostendorf,
Facial
Image Synthesis, Analysis, and Recognition
Demetri Terzopoulos,
Advances
in Scalable Video Compression
Jens-Rainer
Ohm,
Smart
Surveillance: Advanced Video Analytics and Middleware for Security and Retail
Applications
Andrew Senior, IBM
Network Coding for the Internet and Wireless Networks
Philip
Chou, Microsoft
Capturing
and Rendering Three-Dimensional Auditory and Visual Scenes
Ramani Duraiswami,
When:
Where: Crystal
Ballroom
Title:
Managing Spoken Documents
Speaker: Mari
Ostendorf
Abstract:
As storage costs drop and
bandwidth increases, there has been a rapid growth of information available via
the web or in online archives, raising problems of finding and interpreting
collections of documents. Significant recent progress has been made in text
retrieval, analysis, summarization and translation, but much of this work has
focused on written language. Increasingly, speech and video signals are
also available -- including TV and radio broadcasts, congressional records,
oral histories, voicemail, call center recordings, etc. -- which can be thought
of as ``spoken documents''. Because it takes longer to listen to audio
than to read text, spoken documents are clearly a prime candidate for automatic
indexing, information extraction, and other such technologies. In this
talk we overview speech processing technology that underlies spoken document
management, including mathematical frameworks for both word and metadata recognition,
and for integrating video and language cues. In addition, we discuss
issues that arise in text processing when moving from written to spoken
language and implications for statistical models of language.
Outline:
1. The
role of speech in multimedia
- Driver: the
problem of managing unstructured information
- Enablers:
(a) Advances in
text processing
(b) Advances in
speech and speaker recognition
2.
Application challenges in processing speech vs. text
- Speech recognition
errors
- Conversational
speech issues (disfluencies, wording)
- Opportunities in combining
speech with video
3.
Mathematical models of speech and language
- Key issues
(a) Discriminative vs. generative modeling
(b) Symbolic and/or continuous observations
(c) Supervised vs. unsupervised learning
- Key modeling problems
(a) Categorization
(b) Sequence labeling
(c) Sequence transformation
4.
Challenges of "real" systems: uncertainty propagation, dynamic nature
of language, etc.
5.
From speech to video: challenges and opportunities
When:
Where: Crystal
Ballroom
Title: Facial Image Synthesis, Analysis, and
Recognition
Speaker: Demetri Terzopoulos
Abstract:
Methodologies spanning computer graphics, computer
vision, and machine learning are of significant interest in multimedia signal
processing. In particular, we overview several model-based and image-based
methods for synthesizing, analyzing, and recognizing facial imagery. The
model-based methods include functional models of the human face/head, which
incorporates biomechanical synthetic tissues with embedded muscle actuators,
and techniques for applying them to computer animation and expression
estimation in video. The image-based methods include new representations of
facial image ensembles that disentangles pose, illumination, and expression
effects to improve facial image compression and recognition. To this end, we
present nonlinear generalizations of principal components analysis (PCA) and independent
components analysis (
Outline:
1. Introduction and motivation
2. Facial image synthesis
-
Model-based methods: Geometric and biomechanical facial models
- Image-based methods: Video-based synthesis of facial animation
3. Facial image analysis
-
Model-based methods: Expression estimation
- Image-based methods: Facial ensemble analysis and learning
4. Facial image recognition
-
Model-based methods: Morphable models
- Image-based methods: Eigenfaces and TensorFaces
5. Research directions
When:
Where:
Crystal
Ballroom
Title:
Advances in Scalable Video Compression
Speaker: Jens-Rainer Ohm
Institute of Communications Engineering
ohm@ient.rwth-aachen.de
Abstract:
The market for digital video is growing, and compression technology is a
core enabling technology. However, the interrelationship between transmission
networks and compression technology bears many problems yet to be solved. It is
well known that efficient scalable representation of video would be useful to
provide flexible multi-dimensional resolution adaptation, to support various
network and terminal capabilities, and provide better error robustness.
Scalability in traditional hybrid video coding has so far been mainly disabled
by the drift problem, which occurs when different predictions are made at the
encoder and decoder sides. It imposes severe constraints in particular when
different dimensions of scalability shall be supported, e.g. to enable changes
in temporal and spatial resolution as well as fidelity of quantization. Recent
advances in motion-compensated interframe operation
include closed-loop prediction with limited drift, and fully open-loop temporal
compression methods, such as motion-compensated temporal filtering (MCTF), a
framework that extends traditional interframe
compression into more general temporal-axis transform schemes. This finally
turns efficient scalable video compression into reality. The talk will give an
in-depth analysis of the current trends and report about the recent scalability
extension of the Advanced Video Coding (AVC/H.264) standard. The new
developments for scalable compression have provided a deeper insight into
benefits that can be taken from spatio-temporal
coherences over shorter and longer distances. Therefore, some of the new
compression tools even have the potential to further improve the compression
efficiency. On this basis, the talk will give an outlook discussing potential
future trends in video compression, which appears still far from reaching its
final bounds.
Outline:
1.
Advantages of scalable video coding (SVC) from an application perspective
2. An
analysis why scalable video compression did not work efficiently in traditional
motion compensated coding schemes
3. Recent
developments of motion-compensated temporal filtering, open-loop compression
and drift control schemes that enable scalable video coding with high
compression efficiency
4. An
overview about the present SVC standardization activities in MPEG and JV
5. An
outlook to potential future improvements
When:
Where: Crystal Ballroom
Title:
Smart Surveillance: Advanced Video Analytics and Middleware for Security and
Retail Applications
Speaker: Andrew
Senior
IBM
Abstract:
Video surveillance is a
burgeoning field in computer vision. Terrorism has spurred investment in video surveillance
and demands for increased scalability coupled with increased computing power
have led to a huge growth in development of automated surveillance software
that can "watch" surveillance video instead of fallible security
guards. In this tutorial, I will cover the technical computer vision techniques
that enable the automation of video surveillance, together with the scalability
and data management issues that are required for the deployment of real
"smart surveillance" applications, and will end by discussing
important issues of performance analysis and privacy.
Outline:
1.
Challenges of automated video surveillance
(a) Detection:
- Background subtraction techniques
- Alternative detection strategies
(b) Tracking
- Goals
- Techniques
2. Data
storage
(a)
Metadata extraction
(b)
Video storage
3.
Scalability
(a) Distributed systems
(b) Edge devices and smart cameras
4.
Building smart surveillance applications
(a) Smart surveillance capabilities: alerts,
search
(b) Applications: security systems, retail and
other
5.
Performance Analysis
(a) BGS performance
(b) Tracking performance
(c) Task performance
6.
Privacy
(a) Privacy-protecting surveillance
(b) Surveillance privacy certification
When:
Where: Crystal Ballroom
Title: Network Coding for the Internet and Wireless Networks
Speaker: Philip Chou
Microsoft Research
pachou@microsoft.com
Abstract:
In a network, imagine
a pair of 100-Kbps streams arriving at a node from different directions and competing
to be forwarded out of the node on a single 100-Kbps bottleneck link. How
can both streams be forwarded through the bottleneck? The answer is:
combine the two bit streams by superimposing them as vectors over a finite
field, send the combined stream through the bottleneck link at 100 Kbps, and
finally untangle the streams at their destinations, using side information.
This is an example of Network Coding, in which a small amount of linear
processing at some of the nodes in a network can yield increases in throughput,
decreases in delay, and decreases in energy (in a wireless network), relative
to what can be achieved if only pure forwarding (no coding) is allowed.
In this talk I will give an introduction to this new field in practical terms:
how to improve communication in the Internet and wireless networks.
Outline:
1. Introduction to network coding theory
2. How
to make network coding practical
3.
Internet and wireless applications
a)
Live broadcasting,
b)
File downloading
c)
Storage
d)
Messaging
e)
Interactive communication
f) Sensor networks
When:
Where: Crystal Ballroom
Title: Capturing and Rendering Three-Dimensional Auditory and Visual Scenes
Speaker: Ramani Duraiswami