The Symposium on Subbands and Wavelets was held at NJIT in Newark as a One Day Symposium in March 1994, as the third in a series. Fourteen speakers gave talks on digital audio broadcasting, spread spectrum communications, image compression, time-frequency analysis of short pulse scattering data, disparity estimation for stereo image compression, and video compression coding. Twelve of these papers are included in the Video Proceedings.
List of Included Papers:
|
L. Hinderks, Digital Audio Broadcasting: Current Status and Future Directions, Corporate Computer Systems, Inc. |
|
M.J. Medley, P.K. Das, and G.J. Saulnier, Applications of Discrete Wavelet Transform Excision to Spread Spectrum Signals, RPI |
|
G. Schuller and M.J.T. Smith, A General Formulation for Modulated Perfect Reconstruction Filter Banks with Variable System Delay, Georgia Tech |
|
R.A. Haddad and K. Park, Optimum Design of Quantized M-Band Filter Ganks, Polytechnic University |
|
P. Moulin, Optimization Techniques for Efficient Image Compression in Nonorthogonal Multiresolution Bases, Bellcore |
|
A. Benyassine and A.N. Akansu, Subspectral Modeling in Filter Banks,, NJIT |
|
L. Carin, L.B. Felsen, D. Kralji, S.U. Pillai, and M.S. Oh, Time-Frequency Analysis of Short-Pulse Scattering Data: STFT, Wavelet, and ARMA Processing, Polytechnic University |
|
S. Sethuraman, A.G. Jordan, and M.W. Siegel, Multiresolution Based Hierarchical Disparity Estimation for Stereo Image Pair Compression, Carnegie Mellon University |
|
R. Sefranek, Perceptually Based Prequantization for Image Compression, AT&T Bell Laboratories |
|
C.F. Barnes, J.S. Goldstein, M.A. Ingram, and E.J. Holder, Stochastis Successive Approximation Quantization of Image Subbands, Georgia Tech Research Institute |
|
D. Cooper, R. Buell, M. Sherman, and A. Akansu, Adaptive Subband Coding with Multiplierless Filters, GEC-Marconi and NJIT |
|
J. Wus, W. Li, and Y.-Q. Zhang, New Vector Subband Coding for Image and Video Compression, Lehigh University and GTE Laboratories |
plus
by Prof. Mark J. T. Smith, School of Electrical Engineering,
Georgia Tech
Dr. Henrique Malvar, Chief Scientist, PictureTel Inc.
an introductory course on the principles of three dimensional (3D) optical microscopy
|
Collecting 3D data from a microscope, including optical sectioning methods using the widefield fluoresence and transmitted-light-brightfield microscopes, and including the confocal fluoresence microscope. |
|
Modeling of the blur characteristics of the microscope, including the concepts of the optical transfer function (OTF) and the point spread function (PSF). |
|
The theory of computational deblurring, with an emphasis on the maximum-likelihood based algorithms that the lecturer has developed. Deblurring algorithms are used to restore sharpness and clarity to, in effect, remove the haze and blur that is due to intervening out-of-focus structures. |
|
Blind deconvolution methods, where the PSF of the microscope does not have to be measured. Instead, the PSF is reconstructed computationally from the same noisy and blurred image data from which the image is being restored. |
|
Example image reconstructions are shown. |
Biologists, clinicians, engineers, and programmers interested in the crirical elements of 3D microscopy systems. This video course will give you a basic understanding of the capabilities of 3D microscopy systems.
The two video cassettes (2 1/2 hours) are accompanied by copies of 35 visuals, including conceptual schematics of the microscope system, optical sectioning data collection, deblurring models, and basic portions of the computational algorithms.
Professor Timothy J. Holmes (DSc Washington University, St. Louis, 1985) has made a number of recognized contributions to 3D optical microscopy. These include the adaptation of the maximum-likelihood principle to 3D microscopy, which had earlier been applied to nuclear medicine imaging, the introduction of a blind deconvolution algorithm, and simple practical means for reconstructing 3D images of absorbing stains using an ordinary transmitted light brightfield microscope. In 1985 he joined the Electrical and Computer Engineering faculty at the University of Missouri - Columbia and in 1989 he transfered to the Biomedical Engineering Department at Rensselaer Polytechnic Institute in New York State.
BACKGROUND: Optical sectioning, Microscope/computer system, Widefield fluoresence, Transmitted light brightfield and confocal fluoresence modalities.
MODELING: PSF, OTF, and noise models, Maximum-likelihood principle, Purpose of deblurring and blind deconvolution.
DEBLURRING ALGORITHM: Basic algorithm flowchart, Mathematical constraints on the PSF solution.
EXAMPLE IMAGE RECONSTRUCTIONS Basic 3D rendering principles, 2D fluoresence resolution improvement, 3D fluoresence example, Confocal fluoresence example, Typical computer processing times.
|
some of the most significant approaches that have been investigated since the first patents were files at the begining of the 19th Century; |
|
the immense potential range of applications of high-performance analysis and optical character recognition; |
|
fast preprocessing techniques for noise and skew removal, and for the identification of column and line structure; |
|
what types of features have proved most successful for multifont and handprinted characters; |
|
why common statistical assumptions fail, and the strengths and weaknesses of different classification methods; |
|
how methods based on lexicons and grammars, and the stylistic uniformity of entire passages of text, can be used to increase the accuracy of the conversion process; |
|
who are the major vendors of OCR equipment, and how do their products perform; |
|
what software and data are most useful for developing and evaluating new systems and where to obtain additional information. |
Researchers and developers in the areas of; document analysis sorting; pattern recognition, image processing, and information retrieval courses; managers and engineers of projects for developing intelligent character recognition and document image handling systems; system integrators; moderate to large scale OCR users, including librarians, information scientists, and service bureaus for data conversion.
The video cassettes are accompanied by 400+ visuals, including illustrations of both the input and output of commercial and research systems; lists of widely used preprocessing algorithms, features, classifiers and commercial OCR devices; examples of linguistic and spatial context; relevant books, conferences, bibliographies, and test databases; several hundred references from George Nagy's recent papers; and a Who is Who.
Professor George Nagy (PhD Cornell 1962) has made a number of recognized contributions to pattern recognition and OCR research in both industrical and academic settings. He has consulted with some of the leading producers of OCR equipment and has held visiting appointments at several universities and research laboratories. Many of his papers on the subject have been reprinted and widely cited, and he has presented over one hundred talks on related topics at conferences, universities, and industrial organizations. Since 1985 he has been Professor of Computer Engineering at Rensselaer Polytechnic Institute.
INTRODUCTION: History, Applications, Taxonomy
PROCESSING: Scanners, Binarization, Noise removal, Skew removal, Image storage formats
FORMAT ANALYSIS: Manual zoning, Generic typesetting conventions, Publication-specific techniques
FEATURES: Preclassifiers, Templates, Similarity invariants, Orthogonal decompositions, Structural features, Topological invariants, Feature selection and evaluation
CLASSIFICATION: Bayes' formula, Statistical independence, Parameter estimation, Decision Theory: loss and risk, Popular statistical classifiers, Maximum Likelihood, Non-parametric methods, Decision trees, Unsupervised learning and clustering
LINGUISTIC CONTEXT: Letter n-grams, Markov chains, Lexical methods, Lexicon organization, Approximate matching, Syntax, Semantics, Pragmatics
SPATIAL CONTEXT: Typesetting practics, Writing styles, Reject recovery, Adaptive classification, Combined methods
COMMERCIAL SYSTEMS: Vendors, Software or hardware, Error metrics, Performance evaluation, Current performance levels, Sources of errors
R & D PRIMER: Summary, Toolkit, Homilies, Test databases, Sources of additional information, conferences, special issues, books, bibliographies, references, who-is-who
by William A. Pearlman
|
the principles of image compression by vector quantization (VQ); |
|
how the various vector quantization techniques compare in performance; |
|
methods of using training data to design VQ codebooks; |
|
algorithms to distribute bits optimally among image subbands; |
|
how to utilize VQ in a subband or wavelet image coding system; |
|
the difference between minimum distortion fixed level, and entropy-constrained optimality criteria; |
|
the practical advantages and disadvantages of tree-structured VQ; |
|
how to prune a tree-structured VQ to find the best sub-tree to maximize performance; |
|
how to put entropy constraints into a fast pairwise
nearest neighbor PNN codebook design. Who should see this tutorial:Managers and engineers of projects for developing both software and hardware-based image compression systems; research personnel on projects for image and video coding; faculty in charge of digital image processing, coding, and signal processing courses; digital imaging and systems engineers who want to learn the range of vector quantization techniques from the fundamentals to the forefront of digital imaging research. Printed material:The three video cassettes are accompanied by copies of some 100 visuals, including illustrations of vector quantization and image coding systems. Course Lecturer:Professor William A. Pearlman (PhD Stanford 1974) has made a number of recognized contributions to image and source coding research. His lecture style is very clear and to the point. He is a fellow of the Society of Photo Instrumentation Engineers (SPIE) and a Senior Member of the IEEE. He was conference chairman of the 1989 SPIE Communications and Image Processing Conference. In the Spring of 1993, he was Visiting Professor and Lady Davis Scholar at the Technion in Israel. Since 1979 he has been at the Electrical, Computer, and Systems Engineering Department at Rensselaer Polytechnic Institute, where he is now Professor. Video Course Outline1. Vector Quantization (VQ) System 2. Complexity and Distortion Criteria 3. Minimum Distortion Optimality Criterion 4. Generalized Lloyd Algorithm (GLA or LBG) 5. Entropy-Constrainted Optimality Criterion 6. Entropy-Constrained (ECVQ) Design Algorithm 7. Pairwise Nearest Neighbor (PNN) Codebook Design - (i) minimum distortion, (ii) entropy-constrained (ECPNN) 8. Alphabet Constrained VQ Design - (i) alphabet constrained ECPNN, (ii) alphabet constrained ECVQ 9. Performance Comparisons - Gaussian sources, squared error 10. Tree-Structured VQ - (i) balanced trees, (ii) unbalanced trees, (iii) pruning, (iv) comparative simulations 11. Subband Decomposition and Coding - (i) subband system, (ii) types of decompositions for images: QMF, wavelet, rectangular, hexagonal, (iii) rate allocation by BFOS algorithm, (iv) image simulation results for AECVQ, ECPNN, and AECPNN. Vector Quantization Techniques - f.o.b. US$ 465
|
