[Home] [Back Issues] [Search] [Vol 1] [Vol 2] [Vol 3] [Vol 4] [Vol 5]

The Wavelet Digest - Volume 4, Issue 4 (April 12, 1995)


Today's Editor: Wim Sweldens
                sweldens@math.sc.edu


Today's Topics:

     1. Preprint: Adaptive hierarchical approximation on the sphere
     2. Preprint: Characterization of signals by the ridges ...
     3. Preprint: Wavelet correlations in hierarchical branching processes
     4. Preprint: Preprints from the IRISA ftp host (A. Juditsky)
     5. Preprint: Spherical wavelets: Texture processing
     6. Preprint: CREW: Compression with Reversible Embedded Wavelets
     7. Software: A fast matlab routine for calculating Daubechies filters
     8. Software: Application of the Joint Time-Frequency Analyzer
     9. Software: Discrete wavelet transform analysis (WPLW)
    10. Meeting:  Fractal Image Encoding and Analysis ASI (CFP)
    11. Course:   The 5th international summer school
    12. Course:   Mathematics and Physics of Wavelets with Applications
    13. Answer:   Wavelets and convolution (WD 4.2 #28)
    14. Question: Derivatives of wavelets
    15. Question: Identification of frequency modulation laws.
    16. Question: Wavelets and speech recognition.
    17. Question: Reference missing in Daubechies' Ten lectures ?
    18. Question: Denoising Via Wavelet Techniques
    19. Question: Example 3.2 in Chui's book.
    20. Question: wavelets and space-time data.
    21. Question: FWT from Numerical Recipes
    22. Question: Backend Compression Methods for Wavelet Transformed Images ?
    23. Question: self-similar sum of arbitrary function
    24. Question: spectrum of wavelet opertors
    25. Question: Structure detection, wavelets, preprint?


Submissions:
  E-mail to wavelet@math.sc.edu with "submit" as subject.

Subscriptions:
  E-mail to wavelet@math.sc.edu with "subscribe" as subject.
  To unsubscribe, e-mail with "unsubscribe" followed by your e-mail
  address as subject. To change address, unsubscribe and resubscribe.

Preprints, references, and back issues can be obtained from our
information servers:
  Ftp:    ftp.math.sc.edu (/pub/wavelet)
  Gopher: gopher.math.sc.edu
  WWW:    http://www.math.sc.edu/~wavelet/


Current number of subscribers: 4499

Calendar of events:

 Jan-Apr:       Spline Functions and the theory of Wavelets, Montreal
 Mar  7-Apr 18: Wavelets and Applications Course, Boston
 Apr 17-21:     SPIE: Wavelet applications for dual use, Orlando
 May 17-20:     Course: Wavelets: Principles, Applic. and Implement.
 May 22-24:     Course: Fuzzy Logic, Chaos, and Neural Networks, UCLA
 May 30-Jun 3:  Meeting of the Acoustical Society of America, DC
*Jun  6-8 :     Mathematics and Physics of Wavelets, College Park MD
 Jun 26-30:     ANU Wavelets Workshop, Canberra, Australia
 Jul  3-7 :     SIAM ICIAM 95, Hamburg, Germany
*Jul  8-17:     Fractal Image Encoding and Analysis, Trontheim, Norway
 Jul 13-14:     SPIE: Mathematical Imaging: San Diego
 Jul 24-28:     Wavelets in Electromagnetics PIERS, Seattle
 Jul 31-Aug 25: International Summer School, Jyvaskyla, Finland
 Aug 31-Sep 1:  UK Symp. on Time-Freq. and Time-Scale, Warwick UK
 Sep 17-21:     ASME Wavelets in Vibrations and Acoustics, Boston
 Dec 10-13:     Neural Networks and Signal Processing, Nanjing, China


From: roehlich@zib-berlin.de (Jochen Froehlich) Subject: Preprint: Adaptive hierarchical approximation on the sphere Title: An adaptive hierarchical approximation method on the sphere using axisymmetric locally supported basis functions Authors: R\"udiger Brand, Willi Freeden and Jochen Fr\"ohlich Abstract: Lacunary hierarchical approximation in a wavelet-type basis has to cope with the discretization problem for the computation of the retained amplitudes. Furthermore, in sperical geometry no uniform discrete shift operator is available. The paper describes a hierarchical approximation method which has been constructed in the multiresolution spirit using non-orthogonal bump-like basis functions. It involves discrete shift, dilation and scale separation only in a qualitative way. The resulting scheme is appealing by its simplicity and nevertheless successfull and efficient for real life scattered data approximation on the sphere. University Kaiserslautern, Math Dep., Preprint 130. Konrad-Zuse-Zentrum Berlin, Preprint SC 95-2 anonymous ftp: ftp://ftp.ZIB-Berlin.DE/pub/zib-publications/reports/ www: http://www.zib-berlin.de/ZIBbib/Publications/
From: Rene Carmona <rcarmona@chelsea.math.uci.edu> Subject: Preprint: Characterization of signals by the ridges ... Preprint Available by anonymous ftp from chelsea.math.uci.edu Title: Characterization of signals by the ridges of their wavelet transforms Authors: Rene A. Carmona, Wen L. Hwang and Bruno Torresani Abstract: Many procedures have been developed to characterize a signal by some of the salient features of a specific transform, whether it is the Fourier, the Gabor, the Wigner or more recently the wavelet transform. Ridges in the modulus of the transform determine regions in the transform domain with a high concentration of energy. For this reason they are regarded as natural candidates for the characterization and the reconstruction of the original signal. We present a couple of new algorithmic procedures for the detection of ridges in the modulus of the (continuous) wavelet transform of one-dimensional signals. These detection procedures are shown to be robust to additive noise. We also derive and test a new reconstruction procedure. The latter uses only information of the restriction of the wavelet transform to a sample of points from the ridge. This provides with a very efficient way to code the information contained in the signal.
From: 'Peter Lipa' <Peter.Lipa@oeaw.ac.at> Subject: Preprint: Wavelet correlations in hierarchical branching processes Title: Wavelet correlations in hierarchical branching processes Authors: Martin Greiner, Jens Giesemann, Peter Lipa and Peter Carruthers Comments: 31 pages, latex, 10 figures (available from authors). Report-No: HEPHY-PUB 618/95, AZPH-TH/95-03 Abstract: A study of correlations in tractable multiparticle cascade models in terms of wavelets reveals many promising features. The selfsimilar construction of the wavelet basis functions and their multiscale localization properties provide a new approach to the statistical analysis and analytical control of hierarchically organized branching processes. The exact analytical solution of several discrete models shows that the wavelet transformation supresses redundancy in the correlation information. Wavelet correlations can be naturally interpreted as correlations between structures (clumps) living on different scales. Note: This paper adresses to people interested in applications of wavelets to stochastic (cascade) processes, statistical physics, turbulence, high energy physics .... The paper can be obtained electronically from the Los Alamos preprint database: hep-ph@xxx.lanl.gov paper: hep-ph/9503235 (no figures though) The figures can be obtained in an uu-encoded ps-file (1.4 Mbyte !!) via e-mail (or a hardcopy can be sent by snail-mail). Some figures don't print on a NeXT machine. Send e-mail request to: lipa@hephy.oeaw.ac.at Sincerely, Peter Lipa
From: Anatoli Iouditski <Anatoli.Iouditski@irisa.fr> Subject: Preprint: Preprints from the IRISA ftp host (A. Juditsky) title: Wavelet Estimators, Global Error Measures Revisited author: Bernard Delyon, Anatoli Juditsky Abstract: In the paper minimax rates of convergence for wavelet estimators are studied. For the problems of density estimation and non-parametric regression we establish upper bounds over a large range of functional classes and global error measures. The constructed estimate is simultaneously minimax (up to constant) for all L_pi- error measures, 0<pi<=infinity. ftp: irisa.irisa.fr: /techreports/1993/PI-782.ps.Z title: Wavelet Estimators: Adapting to Unknown Smoothness author: Anatoli Juditsky Abstract: A wavelet thresholding algorithm is used to recover a function of unknown smoothness from noisy data. It is known that it can be tuned to be minimax in order over a wide range of Besov-type smoothness constraints and L_p-losses. We provide a method to estimate an adaptive threshold parameter for each resolution level. It is shown that the proposed algorithm is {\em adaptive in order}, i.e. it attains the rate of convergence which is minimax up to a constant over Besov regularity classes and L_p-error measures, 1<=p<= infinity. The algorithm is computationally straightforward: the whole effort to compute the threshold is order N log(N) for the sample size N. ftp: irisa.irisa.fr: /techreports/1994/PI-815.ps.Z title: On the Computation of Wavelet Coefficients author: Bernard Delyon, Anatoli Juditsky Abstract: We consider fast algorithms of wavelet decomposition of function f when discrete observations of f (supp(f) belongs to [0,1]) are available. The properties of the algorithms are studied for three types of observation design: the regular design, when the observations f(x_i) are taken on the regular grid x_i=i/N, i=1, ..., N; the case of gittered regular grid, when it is only known that for all 0< i< N+1, i/N<= x_i<(i+1)/N; the random design case: x_i, i=1, ..., N are independent and identically distributed random variables on [0,1]. We show that these algorithms are in certain sense efficient when the accuracy of approximation is concerned. The proposed algorithms are computationally straightforward: the whole effort to compute the decomposition is order N for the sample size N. ftp: irisa.irisa.fr: /techreports/1994/PI-856.ps.Z title: Computing wavelet density estimator for stochastic processes author: Anatoli Juditsky, Frederique Leblanc Abstract: Let (X_t) be a stictly stationary stochastic process (in continuous or discrete time). We are to estimate the density f of X_t on the basis of discrete observations (X_i), i=1,..., N using a "linear" wavelet estimator. For the continuous time process (X_t), 0<= t<= T those observations are the result of a regular discretization of the continuous time trajectory. We provide an adaptive version of the algorithm, which adapts automatically to the regularity parameters of the density f. To compute the estimator we implement a fast algorithm (the whole effort to compute the estimate is order N log(N) for the sample size N).
From: Wim Sweldens, sweldens@math.sc.edu Subject: Preprint: Spherical wavelets: Texture processing Title: Spherical Wavelets: Texture Processing Authors: Peter Schröder and Wim Sweldens Abstract: Wavelets are a powerful tool for planar image processing. The resulting algorithms are straightforward, fast, and efficient. With the recently developed spherical wavelets this framework can be transposed to spherical textures. We describe a class of processing operators which are diagonal in the wavelet basis and which can be used for smoothing and enhancement. Since the wavelets (filters) are local in space and frequency, complex localized constraints and spatially varying characteristics can be incorporated easily. Examples from environment mapping and the manipulation of topography/bathymetry data are given. Anonymous ftp to ftp.math.sc.edu, file /pub/imi_95/imi95_4.ps or imi95_4.ps.gz. The color images in this paper are available separately in the files /pub/imi_95/imi95_4.pix/*.tif. Also, the postscript file without the figures is available as imi95_4.nofig.ps.
From: schwartz@crc.ricoh.com "Edward L. Schwartz" Subject: Preprint: CREW: Compression with Reversible Embedded Wavelets Title: CREW: Compression with Reversible Embedded Wavelets Authors: Ahmad Zandi, James D. Allen, Edward L. Schwartz, and Martin Boliek RICOH California Research Center Compression with reversible embedded wavelets (CREW) is a unified lossless and lossy continuous-tone still image compression system. It is a wavelet based compression scheme, that uses a "reversible" approximation of one of the best wavelet filters. Reversible wavelets are non-linear filters which implement exact-reconstruction systems with minimal precision integer arithmetic. CREW provides state of the art lossless compression of medical images (> 8-bit deep), and state of the art lossy and lossless compression of 8-bit deep images with a single system. CREW has reasonable software and hardware implementations. Applications for CREW include medical, prepress, satellite and document image compression. Available though WWW at http://www.crc.ricoh.com/misc/crc-publications.html Edward Schwartz RICOH California Research Center 2882 Sand Hill Rd Suite 115 Menlo Park, CA 94025-7022 (415) 496-5712 (415) 854-8740 FAX schwartz@crc.ricoh.com
From: Kevin Amaratunga <kevin@phaeton.mit.edu> Subject: Software: A fast matlab routine for calculating Daubechies filters Here is a matlab routine for computing Daubechies filter coefficients which does the spectral factorization using cepstral analysis. This method works well for long filter lengths and it tends to be faster than methods which make use of root finding algorithms. With the approximately 5.5 CPU seconds on a DECstation 5000/25. A quick check shows that sum(h(k)) = 2.00000000000044 sum(h(k)*g(k)) = 1.122455722880489e-23 sum(h(k)*h(k-2l)) = 1.871421275103674e-13 (Max over l, l != 0) Note that cepstral aliasing due to the FFT is not a significant problem. References: Oppenheim and Schafer - Discrete-Time Signal Processing. ---- % function h = daub(Nh) % % Generate filter coefficients for the Daubechies orthogonal wavelets. % % h = filter coefficients of Daubechies orthonormal compactly supported % wavelets % Nh = length of filter. function h = daub(Nh) K = Nh/2; L = Nh/2; N = 512; % Use a 512 point FFT by default. k = 0:N-1; % Determine samples of the z transform of Mz (= Mz1 Mz2) on the unit circle. % Mz2 = z.^L .* ((1 + z.^(-1)) / 2).^(2*L); z = exp(j*2*pi*k/N); tmp1 = (1 + z.^(-1)) / 2; tmp2 = (1 - z.^(-1)) / 2; Mz1 = zeros(1,N); for l = 0:K-1 Mz1 = Mz1 + binomial(L+l-1,l) * (-z).^l .* tmp2.^(2*l); end Mz1 = 4 * Mz1; % Mz1 has no zeros on the unit circle, so use the complex cepstrum to find % its minimum phase spectral factor. Mz1hat = log(Mz1); m1hat = ifft(Mz1hat); % Real cepstrum of ifft(Mz1). (= cmplx % cepstrum since Mz1 real, +ve.) m1hat(N/2+1:N) = zeros(1,N/2); % Retain just the causal part. m1hat(1) = m1hat(1) / 2; % Value at zero is shared between % the causal and anticausal part. G = exp(fft(m1hat,N)); % Min phase spectral factor of Mz1. % Mz2 has zeros on the unit circle, but its minimum phase spectral factor % is just tmp1.^L. Hz = G .* tmp1.^L; h = real(ifft(Hz)); h = h(1:Nh)'; John R. Williams and Kevin Amaratunga Intelligent Engineering Systems Group Department of Civil Engineering MIT Room 1-250 77 Massachusetts Avenue Cambridge, MA 02139. (617) 253 7657
From: Shie Qian <qian@eagle.natinst.com> Subject: Software: Application of the Joint Time-Frequency Analyzer Application of Time-Frequency Representation for Inverse Synthetic Aperture Radar The response to National InstrumentsU official release, last spring, of the Joint Time-Frequency Analyzer (JTFA) has been tremendous. Hundreds of engineers and scientists around the world now use JTFA in such diverse areas as acoustics, aeronautics, biomedical engineering, communications, economics, and seismology. As we continue to improve the JTFA product (e.g., adding data acquisition functions), we will post some successful application examples, under the permission of authors and related institutions, on the National Instruments ftp site. ftp.natinst.com /support/support_notes/jtfa/issse95/postscript/ or /support/support_notes/jtfa/issse95/mac-viewable/ /support/support_notes/jtfa/issse95/windows-viewable/ The first paper posted, ISSSE95.ps, discusses the application for the inverse synthetic aperture radar. This will be an invited paper for the International Symposium on Signals, Systems, and Electronic T95.
From: ANANDKUMAR <akumar@hns.com> Subject: Software: Discrete wavelet transform analysis (WPLW) A Nice Software package for Discrete Wavelet Transform Analysis (WPLW) I am a researcher, trying to apply Wavelets to compress ECG data. For the past 6 months, I have been working with a package called WPLW (Wavelet Packet Laboratory for Windows) to do my research. I do very strongly recommend this tool for researchers who do not want to write their own code for wavelet analysis. The WPLW is very user friendly, and runs under windows. It has a library of Daubechies, Coifman, and several localized trignometric functions. It is very simple to use, yet very powerful, in that the user can verify multiple transcriptions and the choose the best basis, automatically. (the best basis is chosen based on minimum entropy). The tool gives you the wavelet packet and wavelet transform analysis. I have had very good success with this tool, as I can very quickly do multiple wavelet filter analysis, and the choose the best one that fits the signal of interest. For those of you that are interested in more information, you could contact the distributor A K Peters, Ltd. 289 Linden St. Wellesley, MA 02181-5910 (617) 235-2404 The best part of it all, is the outstanding technical support for this tool, provided by Digital Diagnostic Co. in CT. It continues to be a pleasure dealing with the folks at DDC, who support WPLW. Sincerely ANANDKUMAR J HUGHES NETWORK SYSTEMS GERMANTOWN, MD 20876
From: Frederic Gilbert <Frederic.Gilbert@inria.fr> Subject: Meeting : Fractal Image Encoding and Analysis ASI (CFP) CALL FOR PARTICIPATION Fractal Image Encoding and Analysis a NATO Advanced Study Institute July 8-17, 1995 at the Reso Royal Garden Hotel Trondheim, Norway Objectives The Advanced Study Institute (ASI) is an expository meeting in which invited speakers present new advances in fractal image encoding and analysis. The meeting has two main objectives: First, the ASI will focus on presenting material that is not taught elsewhere and that is not available from any one source. Researchers with fractal image analysis and encoding backgrounds will demonstrate their techniques to each other in order to advance both fields. Workshops in both areas will discuss specific implementation issues and generate a list of interesting unsolved problems. Scope Topics to be covered include, but are not limited to: o Fractal Analysis and Modeling of 2D data. o Multifractal image segmentation o Encoding methods: VQ, classification, breaking the time complexity, search strategies. o Decoding methods: pyramid representation, pixel chasing, finite step decoders, matrix inversion. o Video and Color encoding. o Analysis and Implementation Workshops. Structure/Participation Four invited talks will be given each day, followed by one or two shorter selected talks by participants in the institute. A poster session will be held during the breaks and in the evenings. The institute will also include two workshops, one on image encoding and one on image analysis. Participants wishing to present talks or posters should e-mail, fax, or mail an extended abstract (2-3 pages) describing their proposed lecture/poster-session to the ASI (see contact information) before April 15, 1995. NATO asks that participants stay for the full duration of the ASI. Registration Fees before April 15, 1995 after April 15, 1995 advanced registration rates regular rates regular | US$439.00 | US$589.00 student ** | US$219.00 | US$399.00 ** proof of status required To register, complete the registration form available from the http://inls.ucsd.edu/y/ASI/ WWW page, and return it by mail or fax to (619) 534-7664 with your payment (currently payment can only be made by check or money order made payable to "NATO ASI"; credit card payments will be available soon). Preregister by April 15 to take advantage of advanced registration rates. Registration received after April 15, 1995 will be charged at normal rates. NATO limits the size of the institute to a maximum of 80 people. On-site registration will not be possible. Invited Speakers o THE LIST IS IN THE PROCESS OF BEING FINALIZED. Proceedings Talks given by the invited speakers will be collected in a NATO ASI series book published by Springer Verlag. Participants in the institute will receive a free copy when the book is published. Sponsors o NATO o Rockwell International. o Institute for Nonlinear Science, University of California, San Diego. Scientific and Organizing Committee o Director: o Yuval Fisher o Co-directors: o Jacques Levy-Vehel o Geir Oien o Claude Tricot Contact Information NATO Fractal Image Encoding and Analysis ASI Institute for Nonlinear Science 0402 University of California, San Diego 9500 Gilman Drive La Jolla, CA 92093-0402 USA Phone: (619) 534-5599 Fax: (619) 534-7664 E-mail: asi@poincare.ucsd.edu WWW: http://inls.ucsd.edu/y/ASI/
From: Laura Varpula <iss5@tukki.jyu.fi> Subject: Course: The 5th international summer school The University of Jyvaskyla, Finland, is organizing THE 5th INTERNATIONAL SUMMER SCHOOL 31 Jul - 25 Aug 1995 What is the Summer School? The aim of the International Summer School is to offer advanced courses in various topics to both undergraduate and graduate students. What is the programme of the Summer School? The programme of the 5th International Summer School consists of courses on the following topics: mathematics applied mathematics computer science statistics biology chemistry physics How to apply for the Summer School? To apply for the International Summer School please fill in the application form (http://www.math.jyu.fi/summerschool.html) and send it to the organizers by 31 March 1995. What are the costs? There is no tuition fee for the Summer School. Housing and living expenses and travel costs are the responsibility of each participant. For further information please contact The 5th International Summer School Faculty of Mathematics and Natural Sciences University of Jyvaskyla P.O. Box 35 FIN-40351 Jyvaskyla FINLAND Phone: +358 41 602 205 Fax: +358 41 602 201 E-mail and WWW addresses: iss5@tukki.jyu.fi http://www.math.jyu.fi/summerschool.html Please find enclosed more detailed information about the courses of the Wavelet-topic in the Summer School. MA 1 Introduction to wavelets, 30 h, 31 July - 11 August The topics of the course are: Outline of basic Fou- rier analysis, multiresolution approximation, or- thogonal wavelet bases, wavelet frames, biortho- gonal bases, applications to various fields includ- ing functional analysis. The students are assumed to know the first principles of Lebesgue integration and basic functional analysis. Course book: Y. Meyer, "Wavelets and Operators". Dr. D.L. Salinger, Univ. of Leeds, UK. MA 2 Signal processing and multifractal analysis with wavelets, 30 h, 7 - 18 August The topics of the course are: Signal and image processing with wavelets and other time-frequency algorithms, optimal denoising, wavelets and multi- fractal analysis. The basic understanding of wave- lets is assumed. Course books: Y. Meyer, "Wavelets, Algorithms and Applications" (SIAM 1992); P. Flandrin, "Temps-frequence" (Hermes 1993). Prof Y. Meyer, Universite Paris-Dauphine and the French Academy of Sciences, France. MA 3 Applications of wavelets to analysis, 30 h, 10 - 22 August The topics of the course are: Regularity analysis through wavelets: global analysis related to har- monic analysis (Sobolev spaces, Holder spaces, weighted Lebesgue spaces) and microlocal analysis (with applications to multifractal analysis, partial differential equations and turbulence). Basic know- ledge of Lebesgue integration and basic Fourier and functional analysis is required. Course book: Y. Meyer, "Wavelets and Operators". Also of interest: I. Daubechies, "Ten Lectures on Wavelets" (SIAM, 1992); J.P. Kahane and P.G. Lemarie-Rieusset, "Fourier Series and Wavelets" (Gordon and Breach, to appear). Prof. P.G. Lemarie-Rieusset, Universite Paris-Sud, France.
From: "Jenkins, James W." <JenkiJW1@subtech1.spacenet.jhuapl.edu> Subject: Course: Mathematics and Physics of Wavelets with Applications Offered by The Applied Technology Institute COURSE TITLE: The Mathematics and Physics of Wavelets, with Applications to Remote Sensing INSTRUCTOR: Gerald Kaiser, Mathematical Sciences Department, University of Massachusetts at Lowell TIME & PLACE: June 6-8, 1995 in College Park, MD (301) 345 6700 DESCRIPTION: 12 lectures, each 50 minutes, followed by 10 minutes for questions. The instructor will also be available for extended discussions. A mix of university, goverment and industry participants are expected. LECTURES: 1. Continuous time-frequency analysis 2. Continuous time-scale (wavelet) analysis 3. Time-frequency sampling theorems (Discrete windowed Fourier transforms) 4. Time-scale sampling theorems (Discrete wavelet transforms) 5. Multiresolution analysis and subband filtering 6. Daubechies' wavelet bases 7. Radar and sonar wideband ambiguity functions: A canonical application of time-scale analysis 8. Radar and sonar narrowband ambiguity functions: A canonical application of time-frequency analysis 9. Physical (acoustic and electromagnetic) wavelets: Localized solutions of wave and Maxwell equations 10. The scattering of physical wavelets, based on conformal translformations 11. Emission and reception of physical wavelets: A variational approach to remote sensing 12. Other applications. Participants will be given a set of lecture notes and a copy of the instructor's recent book "A Friendly Guide to Wavelets" (Birkhauser, 1994). To register or to receive a full course catalog phone the Applied Technology Institute at (410) 531-6034 or Fax (410) 531-1013. Enrollment is limited. For more information about the course content, contact kaiserg@woods.uml.edu.
From: tsuhan@big.att.com (Tsuhan Chen) Subject: Answer: Wavelets and convolution (WD 4.2 #28) Dear Elharti Abdelmoula, In response to your question posted in the Feb 1995 issue of the Wavelet Digest, a wavelet convolution theorem was presented in (See pp. 2018-2019): Tsuhan Chen and P. P. Vaidyanathan, "Vector space framework for unification of one- and multidimensional filter bank theory," IEEE Trans. on Signal Processing, vol. 42, no. 8, pp. 2006-2021, August 1994. This wavelet convolution theorem is similar to, although not exactly the same as, that of the DFT/FFT type. Best regards, Tsuhan Chen AT&T Bell Laboratories Visual Communications Research Department 101 Crawfords Corner Road, 4C528 Holmdel, NJ 07733-3030 Tel: (908) 949-2708 Email: tsuhan@research.att.com
From: Tony Cai <tcai@compstat.wharton.upenn.edu> Subject: Question: Derivatives of wavelets I am studying function estimation with discrete observations contaminated by random noise. In my work, I need to know the derivative functions of the father wavelet and mother wavelet (of Daubechies' o.n. family and the Coiflets) . I am looking for a fast algorithm to compute the derivative of the mother wavelet. I wish you can help me. Any help is appreciated. Tony Cai tcai@compstat.wharton.upenn.edu [Answer]
From: "SELEGHIM Paulo 148957" <SELEGHIM@dtp.cea.fr> Subject: Question: Identification of frequency modulation laws. I am looking for references on the identification of frequency modulation laws from the wavelet transform and particularly on the application of the stationary phase method. Keywords are for example instantaneous frequency, ridge and skeleton. I intend to apply these techniques in the study of two-phase flow. Thanks in advance Paulo Seleghim Junior Centre d'Etudes Nucleaires de Grenoble DTP/STI-LEF 17, rue des Martyrs 38054 Grenoble Cedex 9 France EM : SELEGHIM@dtp.cea.fr [Answer]
From: URBINI@aquila.infn.it Subject: Question: Wavelets and speech recognition. Hello, I'm looking for wavelets and speech recognition. I wish info for software (C, C++) or tools for Matlab (for Windows). I'm a student on the University of L' Aquila - Italy. Thanks in advance and sorry for my english... Paolo Urbini URBINI@aquila.infn.it
From: andrewd@ee.uts.edu.au (Andrew Dorrell) Subject: Question: Reference missing in Daubechies' Ten lectures ? Please, Is there a reference missing from Daubechies "Ten Lectures in Wavelets" Chapter 8? Page 282 reads "A first application to image analysis of these biorthogonal bases associated to the laplacian pyramid is given in Antonini et al. (1990)." The references section contains two references bearing Antonini's name but both are 1991 and neither seem directly related. Does anyone know what this reference should be? Will post responses of course. Many thanks, Note from the editor: My guess is that this is a typo and that it should be (1991). Cf. the comment at the start of the bibliography. Mr Andrew Dorrell School of Electrical Engineering, University of Technology, Sydney Phone: 61 2 330 2395 Fax: 61 2 330 2435 email: andrewd@ee.uts.edu.au OR dorrell@ihf.uts.edu.au [Answer]
From: el_williams@ccmail.pnl.gov Subject: Question: Denoising Via Wavelet Techniques Comment: Software: I am an avid user of Wavelet Packet Laboratory for Windows. This is a commercially available package sold via Digital Diagnostic Corporation in Hamden, CT. The software has allowed me to extensive research in signal processing of Mass Spectrometry Data. Question: Need Resources on Denoising Techniques: The following is my current perception. Wavelet Techniques for denoising: I. What is Denoising? A. Coherent structure extraction 1. Let's view our signal of interest as being composed of a noisy incoherent signal plus an information signal. We would really like to capture or extract the information signal and discard the noise. 2. The task is to choose a basis in which to represent the information signal. The basis function must be chosen such that the correlation between it and the information signal is maximized. This will then leave the noisy signal not well correlated with the basis function. Of course now this noisy signal is left susceptible to elimination via some non-linear statistical filter. 3. There is a library of Wavelet bases that one can choose from. How does one determine which basis to choose? One method is to perform an exhaustive search of the library selecting the basis that has the least amount of cost. By cost we mean, which basis represents our signal in the most efficient way. So we just simply select a basis in which are signal has a minimum entropy. Now, the work has been done for us with commercial software packages; all that we have to do is to choose the minimum best basis from the set of best basis derived from the library. The best basis gives the smallest entropy expansion of our signal. We can also view this as the smallest measure of distance between our signal and its orthogonal decomposition on some subspace. 4. Now that we have a nice representation of our signal along the best-basis, select the coefficients above a threshold say, t (like above 0.1 %), an eliminate the others. We then reconstruct the signal from these coefficients and save it. Ah ha! There may be coherent structures still hiding in the residual (i.e. buried in the noise if you will). How do we know when to stop looking for hidden structures in the residual!!?? Well once again we need some sort of threshold to compare against So that we continue to test for information in the residual until we reach a point where the presumed extracted coherent part is greater than some threshold. We can also just choose to iterate a fixed number of times (i.e. the old trial by error method). Now unless you have develop some model for which you think you understand the nature of this noise, then it is time for human interpretation. (i.e. are you happy with the results that you see? Do the results make sense? Are the results useful?). I guess what I am trying to emphasize is that in order to calculate "signal-to-noise" ratios one must be able to measure the noise. Everyone likes to assume that the noise is white noise or guassian. Sometimes the assumption is a good approximation. Anyway enough babbling, just be careful on how you interpret your results. This is not magic you have to be interactive!!!!! 5. Finally, I would to describe another denoising technique using wavelets. This one comes to us via Stanford. A technique known as Wavelet Shrinkage was developed Donoho and Johnstone etc... at Stanford University. Those guys are very good statisticians and you know statisticians are always looking for robustness in both parametric and non-parametric data analysis. These guys view denoising as non-parametric statistical estimation and smoothing, whereas I view these operations as nonlinear filter theory. Same thing right!!???? Well this method starts with the idea that we want to estimate our signal of interest, however, it is contaminated by iid zero mean guassian noise. Then they use the additive noise model. Now the next step is to model your signal of interest if you have a lot of apriori (or any at all) information or you have a nice phenomenological model of your signal (i.e. you what's going on with the physics/chemistry of your process). The latter technique is a parametric approach. If on the other hand you don't know much about your signal then nonparametric methods should be employed. Let's get back to wavelet shrinkage. Well manually it is very simple;since noise effects all wavelet coefficients by shrinking the coefficients toward zero we can reduce the noise and simultaneously preserve the features of our signal. (i.e. blurring of any sharp feature is minimize). Now we simply do the inverse transform and viola!!!! We have a denoised signal. I hope didn't lead you on by saying that this was easy. The hard part is to select a wavelet basis and come up with a statistically based thresholding rule. The thresholding rule is the means by which you decide how much to shrink the wavelet coefficients. There are few thresholding rules developed by Donoho and his crew. Such as , thresholding based on SURE (Stein Unbiased Risk Estimator), minimax criteria, etc... There are details also concerning the type of Shrinkage function. (i.e. hard shrinkage vs. soft shrinkage). Thank You for your help in advance, Ernest L. Williams Jr. Battelle Pacific Northwest Labs Applied Physics Center (509) 375-3930 e-mail el_williams@pnl.gov
From: keith.weintraub@tf-mail.citicorp.com (Keith Weintraub - dpt12) Subject: Question: Example 3.2 in Chui's book. I'm a newcomer to wavelets and I am trying to learn by reading "An Introduction to Wavelets" by Charles K. Chui (1992). I am stuck on the proof of example 3.2 on page 54. I believe the result (which can be arrived at by computing and inspecting the Fourier transforms) but the proof as written in the book escapes me. Any ideas? Also, does anyone know of an errata list for the book? Thanks Keith Weintraub (KW) -- Citibank Global Derivatives 399 Park 7/2, NY, NY 10043 keithw@Citicorp.COM uunet!ccorp!keithw 212-291-5827
From: baert@astr.ucl.ac.be (Etienne Baert) Subject: Question: wavelets and space-time data. I am looking for some informations on the feasibility of using wavelets transform for space-time data analysis. Are they anyone working on this area or aware of people who are working in this area ? Any suggestions on whether it is feasible or issues involved will be appreciated. Please write to me at : baert@astr.ucl.ac.be Thanking you in advance. Baert Etienne Institute of Astronomy and Geophysics Georges Lemaitre Universite Catholique de Louvain Louvain-la-Neuve (Belgium)
From: PSOMMA@chiostro.univr.it Subject: Question: FWT from Numerical Recipes We are using an adapted version of the FWT from "Numerical Recipes in C". We tried many other Daubechies's wavelets, but we had problems with the biorthogonal wavelets, in particular when the number of coefficents is odd. In this case, how should be the wrap- around of the matrix? And its inverse? We will be grateful to anyone that can help us. Paolo Sommaruga, Stefano Lonardi email: PSOMMA@chiostro.univr.it Universita di Verona ITALY
From: sev@gdstech.grumman.com (Sev Binello) Subject: Question: Backend Compression Methods for Wavelet Transformed Images ? Hi, I am experimenting wavelet compression on IR, radar and visual images. I have heard of some large compression ratios available via wavelet compression, however, I do not seem to be able to attain such ratios ( i.e often hear of 100 : 1). Are there benchmarks available anywhere? Are descriptions of the techniques used to compress the images available? Basically, I perform a wavelet transform on an image, and iteratively transform the low-low pass image a variable number of times. Then scale the coeeficients back to pixel values and then use various back end compression methods on the transformed and scaled images. I have tried both lossy (Vq, Adpcm) and lossless (LZW, ARIITHMETIC) techniques, in addition to cascading the two. The compression ratios I get are usually on a par with JPEG if not a little lower. Comments and Suggestions Appreciated Sev
From: Amy Caplan <amyc@well.sf.ca.us> Subject: Question: self-similar sum of arbitrary function I am curious about under what conditions the sum of a translated, scaled "basis" function g() will tend to generate a statistically self-similar function, as k->inf: f(t) = sum_k w[k]*g(s[k]*t-o[k]), where o[k] translates and s[k] and w[k] stretch and scale the function g(). Does anyone have any guidance to offer about how to reason about this? Suppose s[k] are all 1. It seems obvious to me (thinking 'spectrally') that no possible w[k] will generate a self-similar image (unless g() already has a 1/f type spectrum. let's assume it does not). Allow s[k] to change: increasing s[k] will shrink g() and create a proportional bandwidth increase; the peak frequency amplitude will also go down so as to maintain constant "energy". If you plot the spectrum of different g() with s() taking on many values it looks like the envelope of these spectra are already 1/f. But, what is the distribution of s[k] needed to make this true? Intuitively, it seems to me that - the positioning o[k] does not affect the spectrum and can be random. - if w[k] = 1 and s[k] are chosen to be uniformly distributed values, then the spectrum might be 1/f. The function will be 'wierd' however-- isolated copies of highly-shrunken g() will poke up in random places. - if s[k] are chosen proportionally, e.g. twice as many s[k]=2 as s[k]=1 and the corresponding w[k] falloff correspondingly somehow, a standard fractal construction will result? - If so, what should the w[k] be? (1/s[k]??), *and how do you conclude this?* Thanks for any guidance (or discussion!)
From: Jean-Philippe Brunet <brunet@Think.COM> Subject: Question: spectrum of wavelet opertors What is known about the spectrum of wavelet operators? The spectrum of the Fourier transform operator is uniformely distributed on the unit circle. I am curious as to what is the distribution of eigenvalues for a wavelet transform operator, in particular whether it would have a fractal character. I tried a few keyword searches in the Wavelet digest but I thought I would ask you directly. Regards, Jean-Philippe brunet Thinking Machines Corporation 245 First Street Cambridge, MA 02142 I computed the eigenvalues of the discrete wavelet transform matrix (DAUB4) of moderate size (N=1000). The distribution on the unit circle is non uniform, but that's all I can really say. If one were to relax the orthogonality constraint then the eigenvalues will spread over the complex plane, perhaps showing interesting (fractal?) patterns (but are there non othogonal wavelet operators of practical interest?). Before I spend some time on this I was curious of what had been done.
From: berg@pool.informatik.rwth-aachen.de (Stephen R. van den Berg) Subject: Question: Structure detection, wavelets, preprint? I am reading an abstracts pamflet here, it contains the following reference to an article that is yet to be published: E. Lega, H. Scholl, J.-M. Alimi, A. Bijaoui and P. Bury A parallel algorithm for structure detection based on wavelet and segmentation analysis Departerment C.E.R.G.A., Observatoire de la Cote d'Azur, B.P.229, 06304 Nice, France Departerment G.D.Cassini., Observatoire de la Cote d'Azur, B.P.229, 06304 Nice, France Laboratoire d'Astrophysique Extragalactique et de Cosmologie, U.R.A. CNRS 173, Observatoire de Paris-Meudon, 92195 Meudon, France The abstract begins with (no, I'm not going to type it in completely :-): We present a parallel algorithm which allows to recognize rapidly structures in a 3-dimensional set of discrete data points resulting from numerical experiments, and to study their morphological properties. ... And now for the question: does anyone have an email address of one of the authors? Or an ftp address/url for the paper? Any pointers are appreciated. Sincerely, berg@pool.informatik.rwth-aachen.de Stephen R. van den Berg (AKA BuGless).

<webmaster@wavelet.org>