Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. An example © 2002 R. C. Gonzalez & R. E. Woods www.imageprocessingbook.com Digital Image Processing, 2nd ed. • The Welch periodogram of this speech signal © 2002 R. C. Gonzalez & R. E. Woods www.imageprocessingbook.com Digital Image Processing, 2nd ed. www.imageprocessingbook.com Problem • Stationary analysis is only applicable to stationary signals. • Non-stationary signals require analysis techniques that encompass time variation – the short-time Fourier transform – wavelets © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com STFT • Speech is stationary within 20ms windows • Fourier transforms are calculated for short windows • Each transform forms one column of a 3D plot © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. STFT Example © 2002 R. C. Gonzalez & R. E. Woods www.imageprocessingbook.com Digital Image Processing, 2nd ed. www.imageprocessingbook.com STFT Example • Each STFT column represents one block of speech • Each frequency component has equal bandwidth • Each STFT element represents the same time scale and bandwidth as every other component © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Wavelets • Wavelets allow multi-resolution analysis: – That is analysis at different scales or resolution. – Standard Fourier analysis only offers a measure of the frequency content of the whole image and the contribution of the image at a particular frequency. – Wavelet analysis permits us to describe an image in terms of frequency at a position in the image. © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Time-Frequency Floorplan © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Wavelets • Each wavelet component has the same timebandwidth product • Those at low frequencies represent longer periods of time than those at high frequencies • A whole variety of wavelets © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Wavelet decomposition • A wavelet decomposition consists of a scaling function and a series of wavelets • The scaling function provides the coarse scale • The wavelets successively refine the representation © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. Scaling Function © 2002 R. C. Gonzalez & R. E. Woods www.imageprocessingbook.com Digital Image Processing, 2nd ed. www.imageprocessingbook.com Decomposition -1st Wavelet, W(x) © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Decomposition -2nd Wavelet W(2x) © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Decomposition -3rd Wavelet W(4x) © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Decomposition – 4th Wavelet W(8x) © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Decomposition – 5th Wavelet © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. Scaling Function www.imageprocessingbook.com • The four co-efficient scaling function is: x c0 (2 x) c1 2 x 1 c 2 2 x 2 c 2 2 x 3 and the corresponding wavelet is W x c3 (2 x) c 2 2 x 1 c1 2 x 2 c0 2 x 3 © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. Haar Wavelet For example the Haar wavelet is c0 c1 1, c 2 c3 0 x) 1 when 0 x 1, x) 0 otherwise The Wavelet is then W x) 1 when 0 x 0.5, W x) 1 when 0.5 x 1.0 W ( x) 0 otherwise © 2002 R. C. Gonzalez & R. E. Woods www.imageprocessingbook.com Digital Image Processing, 2nd ed. www.imageprocessingbook.com Haar Wavelets • The wavelets are then defined by their coefficients which have to satisfy a set of criteria which are called necessary conditions: c 2, c 2, c c 0, c c n n 2 n n n When n n n2 2n n 2n 1 These are Daubechies wavelets © 2002 R. C. Gonzalez & R. E. Woods For the Haar wavelet c0=c1=1 Length 4 Wavelets must satisfy: c 0 (1 cos( ) sin( )) / 2, c1 (1 cos( ) sin( )) / 2, c 2 (1 cos( ) sin( )) / 2, c3 (1 cos( ) sin( )) / 2 Digital Image Processing, 2nd ed. www.imageprocessingbook.com Gabor Wavelet The Gabor wavelet is essentially a sinewave modulated by a Gaussian envelope 2 gwt e -( jf0t) ((t - t0)/a) e Where f0 is the modulating frequency, t0 indicates position and a controls the width of the Gaussian envelope. In 2D this becomes, gw2 D x, y 1 e -(((x - x0) 2 (y - y0) 2 )/2 ) ( j 2f 0 ( (x - x0)cos (y - y0)sin )) e Where x0, y0 control position f0 controls frequency of modulation along either axis and θ controls the direction or orientation of the wavelet. © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Examples of 1-D Wavelet Transform From Matlab Wavelet Toolbox Documentation © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. 2-D Example From Usevitch (IEEE Sig.Proc. Mag. 9/01) © 2002 R. C. Gonzalez & R. E. Woods www.imageprocessingbook.com Digital Image Processing, 2nd ed. Image Pyramids • Creation by iterated approximations and interpolations (predictions) • The approximation output is taken as the input for the next resolution Level • For a number of levels P, two pyramids: approximation (Gaussian) and prediction residual (Laplacian), are created • approximation filters could be: • averaging • low-pass (Gaussian) © 2002 R. C. Gonzalez & R. E. Woods www.imageprocessingbook.com Digital Image Processing, 2nd ed. www.imageprocessingbook.com Gaussian and Laplacian pyramids Starting image is 512×512; convolution kernel (approximation kernel) is 5×5 Gaussian Different resolutions are appropriate for different image objects (window, vase, flower, etc.) J–3 J–2 J–1 J © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Subband Coding • The image is decomposed into a set of band-limited components (subbands). • Two-channel, perfect reconstruction system; two sets of half-band filters • Analysis: h0 – low-pass and h1 – high-pass and Synthesis: g0 – lowpass and g1 – high-pass © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Brief reminder of z-transform properties Downsampled and upsampled by a factor of two sequences in zdomain [2↓]x(n)=xdown(n)=x(2n) ⇔Xdown (z)=1[X(z1/2)+X(−z1/2)] 2 [2↑]x(n)=xup(x)=⎧⎨x(n/2) n=0,2,4,... ⇔Xup(z)=X(z2) ⎩ 0 n = 1,3,5,.... Downsampling followed by upsampling x(n)=[2↑][2↓]x(n) ⇔ X(z)=1[X(z)+X(−z)] Z−1(X(−z))=(−1)n x(n) © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. © 2002 R. C. Gonzalez & R. E. Woods www.imageprocessingbook.com Digital Image Processing, 2nd ed. Families of half-bank PR filters © 2002 R. C. Gonzalez & R. E. Woods www.imageprocessingbook.com Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Seperable filtering in images © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing Aliasing is presented in the vertical and horizontal detail subbands. It is due to the down-sampling and will be canceled during the reconstruction stage. © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com •The standard Fourier Transform (FT) decomposes the signal into individual frequency components. •The Fourier basis functions are infinite in extent. •FT can never tell when or where a frequency occurs. •Any abrupt changes in time in the input signal f(t) are spread out over the whole frequency axis in the transform output F(ω) and vice versa. •WT uses short window at high frequencies and long window at low frequencies. It can localize abrupt changes in both time and frequency domains. © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing Fig. 7.24 (Con’t) © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods Digital Image Processing, 2nd ed. www.imageprocessingbook.com Chapter 7 Wavelets and Multiresolution Processing © 2002 R. C. Gonzalez & R. E. Woods