Page 30 - untitled

Multimedia Processing and Application Labs (MP
Multimedia Processing and Application Labs (MPALs)ALs)
Over the past few years, the Multimedia Tech- D. Digital watermarking: The technique is cient strategy. We propose an embedded block hash in supporting this challenging research. In the past
nologies Group has achieved accomplishments on useful for digital right management, where robust- searching scheme at the decoder side to achieve mo- year, we have developed a trajectory-based real-time
addressing several key problems and exciting ap- ness is a critical issue affecting the practicability tion estimation for recovery of lost packets, while event detection system. The developed system is
plications, and meanwhile displayed impressive of the watermarking system. However, the major maintaining format compliance and cryptographic able to conduct on-line surveillance and the proto-
teamwork for establishing itself as one of the lead- disadvantage of known watermarking methods is provable security. type system is ready for use. In the future, we shall
ing teams in Taiwan. Our recent research efforts are their limited resistance to extensive geometric at- develop new algorithms which will result in better
B. Available Bandwidth Estimation and
highlighted as follows. tacks. We propose a new robust image watermark- accuracy and efficiency.
Wireless TCP: Available bandwidth is an important
ing scheme that can withstand geometric distortions
I. Multimedia Image and Video Processing factor that can be used to adapt the sending rate to B. Addressing Lighting for Vision Prob-
and WEAs simultaneously. Extensive experimental
network conditions, so that packet loss, caused by lems: Handling lighting is one of the most chal-
results obtained using the standard benchmark (i.e.,
A. Video Retrieval: In content-based video
Stirmark) and thorough comparisons with state congestion, can be significantly reduced before error lenging problems in computer vision. We proposed Research Groups
retrieval area, we have developed a video shot-based
of the art technologies confirm the excellent per- control mechanisms are employed. To this end, we a method, generic intrinsic illumination subspace,
fast video clip retrieval system. We make use of the
formance of our method in improving robustness. propose a one-way delay jitter based scheme, ``Jit- which can reduce lighting effects for objects of the
statistics extracted at the shot level to execute fast
To our knowledge, such thorough evaluations and terPath,'' for available bandwidth estimation without same class (eg. human face). This method learns a
search. The results of this work has been published
comparisons have not been reported in the literature relying on common assumptions, including uses of low-dimension subspace of the general appearance
by IEEE Transactions on Circuits and Systems for
before. In addition, we have proposed an asymmet- the fluid traffic model and bottleneck link capac- space formed by images of all the lighting condi-
Video Technology (May issue, 2006). In addition to
ric watermarking method, and demonstrated its se- ity. Extensive simulations and Internet experiments tions of pose-fixed objects of the same class. Then,
the above work, we also calculate the trajectory of
Research Groups
curity and robustness for some attacks. Based on the have been conducted, and comparisons with other this subspace is applied for lighting normalization.
a moving object at the shot level, and then use the
method, we have developed a number of application methods have been made to verify the effectiveness The result was published in one of the most impor-
trajectory to perform fast video retrieval. The devel-
algorithms. of our method. Our ongoing work is to distinguish tant conferences in computer vision, ICCV 2005. In
oped technique can be applied to general monitor-
congestion loss based on wireless loss based on the addition, we introduced a method that can do photo-
ing systems. The core technology developed by our
E. Perceptual Hashing: Perceptual hashing or relationship between available bandwidth and prob- metric stereo under general lighting conditions with
team has been transferred to the Advanced Technol-
digital fingerprinting has been recognized as an al- ing rate to develop a reliable wireless TCP. only four images.
ogy Center (ATC) of ITRI in Taiwan. In the future,
ternative approach for many applications previously
we shall put our emphasis on the development of C. Video codec: Conventional video C. Object Tracking: We proposed a method
accomplished with watermarking. In fact, perceptual
more accurate retrieval algorithms. coding standards, such as MPEG-4 and H.264/ that uses object appearance information to assist
hash is a compact representation of media data. The
AVC usually perform motion estimation among object tracking. Object tracking can be treated as a
B. Representation and Retrieval of 3D major disadvantage of the existing media hashing state estimation problem of a dynamic system, and
technologies is their limited resistance to geometric successive frames so that the encoder is typically
Graphical Models: We are proposing a visual more complex than the decoder. However, this particle filtering is a typical method for object track-

attacks. We propose a novel geometric distortion-
salience-guided mesh decomposition strategy. Its kind of architecture is not suitable for some emerg- ing. However, when the state space dimensionality
invariant image hashing scheme, which is capable
main idea is based on a theory of part salience, ing video coding applications that need resource- is high, particle filtering usually results in drifting
of achieving robustness against extensive geometric
originated from cognitive psychology. The theory limited encoders (e.g. video sensor networks and or local-minimum problem in tracking. Our method
distortions (e.g., standard benchmarks, Stirmark3.1).
asserts that the salience of a part is determined by wireless mobile video communications). Based on adds ‘attractors’ in the state space, and uses attrac-
In addition, a sophisticated hash database for error-
three factors: the boundary strength, the degree of the Wyner-Ziv information theorem, distributed tors to assist tracking. We derived a particle fi ltering
resilient and fast matching is constructed. Our future
protrusion, and the relative size of a part. Since the video coding systems (called Wyner-Ziv video co- method that estimates the maximum a posteriori
work is to study its security against forgery attack.
above-mentioned factors are all conceptual, com- (MAP) solution when the transition probability is
dec) shift part of the computational burden from the
putational processes for modeling these factors are assumed to be a mixture distribution. The result was
F. Texture synthesis: We study texture syn- encoder to the decoder and result in a kind of video
required. We have thus developed a systematic way published in one of the most important conferences
thesize problems by analyzing the performance of codec with low-complexity encoder and high-com-
to conduct 3D mesh decomposition based on visual in computer vision, CVPR 2005. We have applied
the patchwork-based algorithm, which has been plexity decoder. However, this new coding paradigm
salience. The paper describing our results of salient this method to 3D hand tracking and lip-contour
used in a wide variety of applications. Based on our still cannot be applied to applications constrained
component decomposition has been accepted by tracking.
analysis, we extend the algorithm to multi-scale, to need both low-complexity encoder and decoder
IEEE Transactions on Multimedia (March 2006).
multi-class and other applications. (e.g. wireless mobile video communications). In
In the future, we plan to apply this technique to ex- view of this, we study a new media hash-based low-
tracting significant components from a 3D mesh for II. Multimedia Networking, Coding, and complexity Wyner-Ziv video codec, where motion-
3D mesh retrieval. Transmission
compensated interpolation/extrapolation and feed-
C. Video compression: We study several cod- A. Error Resilient Video Encryption and back channel are not required.
ing techniques, such as codecs based on matching Transmission: Media encryption technologies ac-
III. Computer Vision and Pattern Recognition
pursuits, 3D wavelet transforms, and traditional tively play the first line of defense in securing the
hybrid motion compensation approaches. We also access of multimedia data. Traditional cryptograph- A. Real-time Event Detection and Analy-
study multimedia transmission problems, such as ic encryption can achieve provable security but is sis: The aim of the project is to design effective
transmitting in a packet loss environment. Our goal unfortunately sensitive to a single bit error, which algorithms to automatically detect, recognize, and
is to improve current techniques, and produce a new will cause an unreliable packet to be dropped to analyze video objects and events. In the past, we
coding representation and techniques for the next create packet loss. In order to achieve robust media have developed a number of computer vision and
generation of coders. encryption, error resilience is considered an effi- pattern recognition techniques that would be useful
20 21