41
Introduction to Gabor Analysis Radu Frunza Fourier Transform and Operators Gabor Frames Lattices Window Functions Gabor Frames Revised Open questions Final Mention Introduction to Gabor Analysis Theoretical and Computational Aspects Radu Frunza Numerical Harmonic Analysis Group under the supervision of Prof. Dr. Hans Georg Feichtinger 30 Oct 2012 Radu Frunza Introduction to Gabor Analysis

Introduction to Gabor Analysis - univie.ac.at · Radu Frunza Introduction to Gabor Analysis. Introduction to Gabor Analysis Radu Frunza Fourier Transform and Operators Gabor Frames

  • Upload
    others

  • View
    50

  • Download
    0

Embed Size (px)

Citation preview

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Introduction to Gabor AnalysisTheoretical and Computational Aspects

Radu Frunza

Numerical Harmonic Analysis Groupunder the supervision of

Prof. Dr. Hans Georg Feichtinger

30 Oct 2012

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Outline

1 Fourier Transform and Operators

2 Gabor Frames

3 Lattices

4 Window Functions

5 Gabor Frames Revised

6 Open questions

7 Final Mention

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Fourier Transform and OperatorsDFT/ iDFT

• Discrete Fourier TransformGiven an input signal f of length L, the DFT of f is avector F of the same length L, with elements:

F (k) :=

L∑n=1

f(n)e−2πi(k−1)(n−1)/L, 1 ≤ k ≤ L (1.1)

• Inverse Discrete Fourier TransformThe inverse DFT is given by:

f(n) =1

L

L∑k=1

F (k)e2πi(k−1)(n−1)/L, 1 ≤ n ≤ L (1.2)

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Fourier Transform and OperatorsFundamental Operators

• TranslationFor x ∈ ZL we define the translation operator Tx by:

(Txf)(t) := f(mod(t− x, L)) (1.3)

• ModulationFor ω ∈ ZL we define the modulation operator Mω by:

(Mωf)(t) := e2πiωt/Lf(t) (1.4)

• Note:Tx is also called a Time shift, and Mω a Frequency shift.TxMω and MωTx are called Time-Frequency shifts.

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Fourier Transform and OperatorsTranslation, Modulation

• In MATLAB

• Tx = rot(eye(L),−x)• Mω = diag(exp(2 ∗ pi ∗ i ∗ ω ∗ (1 : L)/L))

Figure: Translation Figure: Modulation

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Fourier Transform and OperatorsFundamental Operators II

• InvolutionThe involution ∗ is defined by:

f∗(x) := f(−x) (1.5)

• ConvolutionThe convolution of two vectors f, g of length L is thevector f ∗ g defined by:

(f ∗ g)(x) :=

L∑y=1

f(y)g(x− y) (1.6)

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Fourier Transform and OperatorsConvolution

Figure: Random signal Figure: Result after convolution

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Fourier Transform and OperatorsFundamental Operators III

• Inner ProductThe inner product of two vectors f and g of length L isdefined as:

〈f, g〉 :=

L∑k=1

f(k)g(k) (1.7)

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Fourier Transform and OperatorsShort Time Fourier Transform

• STFTThe discrete STFT of a vector f with respect to a windowvector g is given as:

(Vgf)(x, ω) := 〈f,MωTxg〉 , ∀x, ω ∈ ZL (1.8)

• SpectrogramThe spectrogram of f with respect to a window function gsatisfying ‖g‖2 = 1 is defined to be:

SPECgf(x, ω) := |Vgf(x, ω)|2 (1.9)

• Note(s):The spectrogram non-negative, covariant andenergy-preserving.

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Fourier Transform and OperatorsSpectrograms

Figure: Spectrogram 1 Figure: Spectrogram 2

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Gabor FramesFrames

• DefinitionA finite sequence {g1, g2, ..., gN} of vectors gj of length Lis a frame if ∃A,B > 0 such that:

A

L∑k=1

|f(k)|2 ≤N∑j=1

| 〈f, gj〉 |2 ≤ BL∑k=1

|f(k)|2, (2.1)

for all vectors f of length L.

• Note(s):Any two constants A,B satisfying (2.1) are calledframe bounds.We must have N > L elements in the sequence {gj}j=1,N

in order to obtain a frame.

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Gabor FramesAnalysis Operator

• DefinitionFor any subset {gj : j ∈ J} the analysis operator C isgiven by

Cf = {〈f, gj〉 : j ∈ J} (2.2)

• Note(s):C is also called coefficient operator.In MATLAB: C = [g1; g2; . . . gJ ].

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Gabor FramesSynthesis Operator

• DefinitionThe synthesis operator D is defined for a finite sequencec = (cj)j∈J by

Dc =∑j∈J

cjgj (2.3)

• Note(s):D is also called reconstruction operator.In MATLAB: D = C ′.

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Gabor FramesFrame Operator

• DefinitionThe frame operator S is defined by

Sf =∑j∈J〈f, gj〉 gj (2.4)

• Note(s):In MATLAB: S = C ′C = DD′.

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Gabor FramesDual Frame

• DefinitionIf {gj : j ∈ J} is a frame with frame bounds A,B > 0,then

{S−1gj : j ∈ J

}is a frame with frame bounds

B−1, A−1 > 0, the so-called dual frame.

• Note(s):Every f has non-orthogonal expansions:

f =∑j∈J

⟨f, S−1gj

⟩gj =

∑j∈J〈f, gj〉S−1gj (2.5)

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

LatticesDefinition

• DefinitionA lattice Λ ⊆ Z2

L is a subgroup of Z2L. Any lattice can be

written in the form Λ = AZ2L, where A is an invertible

2× 2-matrix over ZL. Lattices in Z2L can be described as:

Λ = {(x, y) ∈ Z2L|(x, y) = (ak + dl, ck + bl), (k, l) ∈ Z2

L}(3.1)

with a, b, c, d ∈ N and

A =

(a dc b

)(3.2)

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

LatticesSeparability

• SeparableA lattice Λ = aZL × bZL E Z2

L, for a, b ∈ N is called aseparable or product lattice. If Λ is separable then thereexists a generating matrix of Λ which is diagonal:

A =

(a 00 b

)(3.3)

• Non-separableThe generating matrix of any (non-separable) lattice canbe expressed in the form:

A =

(a 0s b

), (3.4)

where s ∈ N.

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

LatticesSpyc

Figure: Separable lattice Figure: Quincux lattice

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

LatticesIndexing and generating lattices

• Matrix representation

• Index storage

• Parameter representation

• L,a,b,s• LTFAT∗ parameters: L,a,M,lt

• Generating lattices

• Separable lattices• Non-separable lattices

∗Information on LTFAT tollboxRadu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

LatticesSubgroups

• Counting ProblemFor a given size N and redundancy red, how many latticesare there?

• SeparableSay that the divisors of N ∗ red are {p1, p2, ..., pn}. Thenumber of separable lattices of redundancy red is ≤ n.

• ExamplesFor L = 480, red = 1.5, there are 10 separable lattices.For L = 480, red = 2, there are 20 separable lattices.

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

LatticesSubgroups II

• Non-separable

Say N =∏j p

ajj and N × red = K =

∏j p

bjj . Clearly

0 ≤ bj ≤ 2aj ,∀j. Then we have:

sK(N) =

n∏j=1

spbjj

(pajj ) =

n∏j=1

{pb+1−1p−1 , 0 ≤ b ≤ a

p2a−b+1−1p−1 , a ≤ b ≤ 2a

• ExamplesFor L = 480, red = 1.5, there are 176 non-separablelattices.For L = 480, red = 2, there are 724 non-separablelattices.

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

LatticesAdjoint Lattice

• NotationFor λ = (x, ω) ∈ Z2

L we typically write

π(λ) = TxMω, λ = (x, ω) (3.5)

• DefinitionFor any lattice Λ ∈ Z2

L we define the adjoint lattice Λo as

Λo ={λo ∈ Z2

L : π(λo)π(λ) = π(λ)π(λo),∀λ ∈ Λ}(3.6)

• ApplicationsJanssen representation/test

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

LatticesSpyc II

Figure: Λ and Λo

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Window FunctionsGaussian window

• DefinitionLet

ϕ(x) = 21/4e−πx2

(4.1)

be the Gaussian on R.

• Properties

• Normalized:

‖ϕ‖2 = 1 (4.2)

• Minimizes the uncertainty principle/ provides the optimalresolution in the time-frequency plane.

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Window FunctionsModified Gaussians

• Stretched Gaussian windows

• Stretching operators• In practice: ϕα(x)

• Rotated Gaussian windows

• Rotation operators• Via Hermite∗ functions

∗Eigenfunctions of the Fourier transformRadu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Window FunctionsModified Gaussians II

Figure: Stretched Gaussian Figure: Stretched and RotatedGaussian

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Window FunctionsDual Window

• DefinitionIf {gj : j ∈ J} is a frame with frame operator S, thenγ = S−1g is called the canonical dual window.

Figure: Dual for a = 20, b = 16 Figure: Dual for a = 10, b = 32

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Gabor Frames RevisedClarifications

Schematically the time-frequency analysis of a signal consists ofthree distinct steps:

• A. Analysis: Given a signal or image f , its STFT Vgfwith respect to a suitable window is computed.

• B. Processing: Vgf(x, ω) is transformed into some newfunction F (x, ω).

• C. Synthesis: The processed signal or image is thenreconstructed using the modified inversion formula:

h =∑λ∈Λ

F (λ)π(λ)γ (5.1)

with respect to a suitable synthesis window γ

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Gabor Frames RevisedConcrete example - Digital image compression

• Method• Given an input 2D signal( image) f , decompose it into

simple parts( RGB/Y CbCr).• For each component, compute a 2D DFT.• Truncate the higher frequencies.• Reconstruct.

• Example

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Gabor Frames RevisedQuality Criteria - Frame Bounds

• Recall the definition of a frameA finite sequence {g1, g2, ..., gN} of vectors gj of length Lis a frame if ∃A,B > 0 such that:

A

L∑k=1

|f(k)|2 ≤N∑j=1

| 〈f, gj〉 |2 ≤ BL∑k=1

|f(k)|2,

for all vectors f of length L.

• Quality of a frameThe ratio B/A is a good indicator of how well a vectorcan be reconstructed using (2.5).

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Gabor Frames RevisedQuality Criteria - CN of Frame Operator

• Frame operatorThe Frame operator plays a central role in the synthesis ofthe signal, either via (2.5) or in computing the canonicaldual window for (5.1). Either way, the inverse of the frameoperator is required in the reconstruction.

• Condition numberThe condition number∗ is a good indicator of theinvertibility of the Frame operator.

∗2-normRadu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Gabor Frames RevisedGeometric Criteria

• DescriptionHow to arrange 2-dimensional balls in R2 in the mosteconomical way.

• Packing problemEach point of R2 may not belong to more than one ball.

• Covering problemEach point of R2 belong to at least one ball.

• Lattice Packing/CoveringA packing/covering is a lattice packing/covering for Λ ifit is of the form (Bx(0) + λ)λ∈Λ

• CriteriaLet r be the radius for the densest packing and R be theradius for the thinnest covering.Then q = vol(BR)/vol(Br) is an indicator of the qualityof the system G(Λ, g0).

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Gabor Frames RevisedPacking/Covering I

Figure: Circles, Sep. Lattice

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Gabor Frames RevisedPacking/Covering II

Figure: Circles, Hex. Lattice

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Gabor Frames RevisedPacking/Covering III

Figure: Circles, Sep. Lattice Figure: Ellipses, Sep. Lattice

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Gabor Frames RevisedQuality Criteria - Janssen Test

• DefinitionGiven a lattice Λ and a window function g, the Janssensum is defined as:

s(Λ, g) :=∑λ∈Λo

Vgg(λ) (5.2)

• ExplanationIf g is properly normalized, i.e. ‖g‖ = 1, then Vgg(0) = 1.If s(Λ, g) < 2, the system has good properties.

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Gabor Frames RevisedOther Quality Criteria

• CN of the dual window

• S0∗ norm of the dual window

• Composite quality criteria

∗1-norm of the STFTRadu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Gabor Frames RevisedEnd note

• Equivalence of Quality criteria?

Figure: Comparison of criteria

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Open questions

• Optimal Quality Criteria?

• Efficient computation of dual windows?

• Efficiency of non-separable lattices?

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

Final Mention

I would like to thank:

• Prof. Dr. Hans Georg Feichtinger

• Dr. Maurice de Gosson

• Christoph Wiesmeyr

• Markus Faulhuber

• Peter Sondegaard

• Everyone in the audience: Thank you!

Radu Frunza Introduction to Gabor Analysis

Introductionto GaborAnalysis

Radu Frunza

FourierTransformand Operators

Gabor Frames

Lattices

WindowFunctions

Gabor FramesRevised

Openquestions

Final Mention

References

• Hans G. Feichtinger, Ole Christensen, Stephan Paukner. Gabor Analysis for Imaging. In ”Handbookof Mathematical Methods in Imaging”, Springer Berlin, Vol.3 p.1271-1307, 2011.

• K. Grochenig. Foundations of Time-Frequency Analysis. Appl. Numer. Harmon. Anal. BirkhauserBoston, Boston, MA, 2001.

• J.H. Conway, N.J.A. Sloane. Sphere Packings, Lattices and Groups, Springer, 1999.

• Chuanming Zong. Simultaneous Packing and Covering in the Euclidean Plane, Monatsh. Math. 134,247-255, 2002.

• Hans G. Feichtinger, Thomas Strohmer, Ole Christensen. A Group-theoretical Approach to GaborAnalysis. Opt. Eng., Vol.34 p.1697–1704, 1995.

• H. G. Feichtinger, N. Kaiblinger. 2D-Gabor analysis based on 1D algorithms. In Proc. OEAGM-97(Hallstatt, Austria), 1997.

• H. G. Feichtinger, W. Kozek, P. Prinz, and T. Strohmer. On multidimensional non-separable Gaborexpansions. In Proc. SPIE: Wavelet Applications in Signal and Image Processing IV, August 1996.

• Stephan Paukner. Foundations of Gabor Analysis for Image Processing, Nov 2007.

• A.J.E.M. Janssen. Gabor representation of generalized functions. J. Math. Anal. Appl., 83:377-394,October 1981.

• D. Gabor. Theory of communication. J. IEE, 93(26):429-457, 1946.

Radu Frunza Introduction to Gabor Analysis