Biorthogonal wavelet filter set
[Lo_D,Hi_D,Lo_R,Hi_R] = biorfilt(
[Lo_D1,Hi_D1,Lo_R1,Hi_R1,Lo_D2,Hi_D2,Lo_R2,Hi_R2] = biorfilt(
biorfilt command returns either four or eight filters associated with
[Lo_D,Hi_D,Lo_R,Hi_R] = biorfilt( computes
four filters associated with the biorthogonal wavelet specified by
DF and reconstruction
RF. These filters are
Decomposition low-pass filter
Decomposition high-pass filter
Reconstruction low-pass filter
Reconstruction high-pass filter
= biorfilt( returns
eight filters, the first four associated with the decomposition wavelet,
and the last four associated with the reconstruction wavelet.
It is well known in the subband filtering community that if the same FIR filters are used for reconstruction and decomposition, then symmetry and exact reconstruction are incompatible (except with the Haar wavelet). Therefore, with biorthogonal filters, two wavelets are introduced instead of just one:
One wavelet, , is used in the analysis, and the coefficients of a signal s are
The other wavelet, ψ, is used in the synthesis:
Furthermore, the two wavelets are related by duality in the
as soon as j ≠ j′ or k ≠ k′ and
as soon as k ≠ k′.
It becomes apparent, as A. Cohen pointed out in his thesis (p. 110), that “the useful properties for analysis (e.g., oscillations, null moments) can be concentrated in the function; whereas, the interesting properties for synthesis (regularity) are assigned to the ψ function. The separation of these two tasks proves very useful.”
and ψ can have very different regularity properties, ψ being more regular than .
The , ψ, and ϕ functions are zero outside a segment.
This example shows how to obtain the decomposition (analysis) and reconstruction (synthesis) filters for the
Determine the two scaling and wavelet filters associated with the
wv = 'bior3.5'; [Rf,Df] = biorwavf(wv); [LoD,HiD,LoR,HiR] = biorfilt(Df,Rf);
Plot the filter impulse responses.
subplot(2,2,1) stem(LoD) title(['Dec. lowpass filter ',wv]) subplot(2,2,2) stem(HiD) title(['Dec. highpass filter ',wv]) subplot(2,2,3) stem(LoR) title(['Rec. lowpass filter ',wv]) subplot(2,2,4) stem(HiR) title(['Rec. highpass filter ',wv])
Demonstrate that autocorrelations at even lags are only zero for dual pairs of filters. Examine the autocorrelation sequence for the lowpass decomposition filter.
npad = 2*length(LoD)-1; LoDxcr = fftshift(ifft(abs(fft(LoD,npad)).^2)); lags = -floor(npad/2):floor(npad/2); figure stem(lags,LoDxcr,'markerfacecolor',[0 0 1]) set(gca,'xtick',-10:2:10)
Examine the cross correlation sequence for the lowpass decomposition and synthesis filters. Compare the result with the preceding figure.
npad = 2*length(LoD)-1; xcr = fftshift(ifft(fft(LoD,npad).*conj(fft(LoR,npad)))); lags = -floor(npad/2):floor(npad/2); stem(lags,xcr,'markerfacecolor',[0 0 1]) set(gca,'xtick',-10:2:10)
Compare the transfer functions of the analysis and synthesis scaling and wavelet filters
dftLoD = fft(LoD,64); dftLoD = dftLoD(1:length(dftLoD)/2+1); dftHiD= fft(HiD,64); dftHiD = dftHiD(1:length(dftHiD)/2+1); dftLoR = fft(LoR,64); dftLoR = dftLoR(1:length(dftLoR)/2+1); dftHiR = fft(HiR,64); dftHiR = dftHiR(1:length(dftHiR)/2+1); df = (2*pi)/64; freqvec = 0:df:pi; subplot(2,1,1) plot(freqvec,abs(dftLoD),freqvec,abs(dftHiD),'r') axis tight title('Transfer modulus for dec. filters') subplot(2,1,2) plot(freqvec,abs(dftLoR),freqvec,abs(dftHiR),'r') axis tight title('Transfer modulus for rec. filters')
Cohen, A. (1992), “Ondelettes, analyses multirésolution et traitement numérique du signal,” Ph. D. Thesis, University of Paris IX, DAUPHINE.
Daubechies, I. (1992), Ten lectures on wavelets, CBMS-NSF conference series in applied mathematics. SIAM Ed.