Presentation is loading. Please wait.

Presentation is loading. Please wait.

Introduction to the Curvelet Transform By Zvi Devir and Yanai Shpinner.

Similar presentations


Presentation on theme: "Introduction to the Curvelet Transform By Zvi Devir and Yanai Shpinner."— Presentation transcript:

1 Introduction to the Curvelet Transform By Zvi Devir and Yanai Shpinner

2 Introduction Curvelet Transform is a new multi-scale representation most suitable for objects with curves. Developed by Candès and Donoho (1999). Still not fully matured. Seems promising, however.

3 Approximation Rates Having an object in the domain [0,1]  [0,1], how ‘fast’ can we approximate it using certain system of functions? Using the Fourier Transform: Using the Wavelet Transform: Using the Curvelet Transform:

4 Point and Curve Discontinuities A discontinuity point affects all the Fourier coefficients in the domain. Hence the FT doesn’t handle points discontinuities well. Using wavelets, it affects only a limited number of coefficients. Hence the WT handles point discontinuities well. Discontinuities across a simple curve affect all the wavelets coefficients on the curve. Hence the WT doesn’t handle curves discontinuities well. Curvelets are designed to handle curves using only a small number of coefficients. Hence the CvT handles curve discontinuities well.

5 Curvelet Transform The Curvelet Transform includes four stages: Sub-band decomposition Smooth partitioning Renormalization Ridgelet analysis

6 Sub-band Decomposition Dividing the image into resolution layers. Each layer contains details of different frequencies: P 0 – Low-pass filter.  1,  2, … – Band-pass (high-pass) filters. The original image can be reconstructed from the sub-bands: Energy preservation

7 Sub-band Decomposition

8 Low-pass filter  0 deals with low frequencies near |  |  1. Band-pass filters  2s deals with frequencies near domain |  |  [2 2s, 2 2s+2 ]. Recursive construction –  2s (x) = 2 4s  (2 2s x). The sub-band decomposition is simply applying a convolution operator:

9 Sub-band Decomposition The sub-band decomposition can be approximated using the well known wavelet transform: Using wavelet transform, f is decomposed into S 0, D 1, D 2, D 3, etc. P 0 f is partially constructed from S 0 and D 1, and may include also D 2 and D 3.  s f is constructed from D 2s and D 2s+1.

10 Sub-band Decomposition P 0 f is “smooth” (low-pass), and can be efficiently represented using wavelet base. The discontinuity curves effect the high-pass layers  s f. Can they be represented efficiently? Looking at a small fragment of the curve, it appears as a relatively straight ridge. We will dissect the layer into small partitions.

11 Smooth Partitioning

12 A grid of dyadic squares is defined: Q s – all the dyadic squares of the grid. Let w be a smooth windowing function with ‘main’ support of size 2 -s  2 -s. For each square, w Q is a displacement of w localized near Q. Multiplying  s f with w Q (  Q  Q s ) produces a smooth dissection of the function into ‘squares’.

13 Smooth Partitioning The windowing function w is a nonnegative smooth function. Partition of the energy: The energy of certain pixel (x 1,x 2 ) is divided between all sampling windows of the grid. Example: An indicator of the dyadic square (but not smooth!!). Smooth window function with an extended compact support: Expands the number of coefficients.

14 Smooth Partitioning Partition of the energy: Reconstruction: Parserval relation:

15 Renormalization Renormalization is centering each dyadic square to the unit square [0,1]  [0,1]. For each Q, the operator T Q is defined as: Each square is renormalized:

16 Before the Ridgelet Transform The  s f layer contains objects with frequencies near domain |  |  [2 2s, 2 2s+2 ]. We expect to find ridges with width  2 -2s. Windowing creates ridges of width  2 -2s and length  2 -s. The renormalized ridges has an aspect ratio of width  length 2. We would like to encode those ridges efficiently Using the Ridgelet Transform.

17 The Ridgelet Transform Ridgelet are an orthonormal set {  } for L 2 (  2 ). Developed by Candès and Donoho (1998). Ridge in Square 2-s2-s 2 -2s Ridge in Square 1 2-s2-s It’s Fourier Transform 2s2s Ridgelet Tiling radius 2 s 2 s divisions It’s Fourier TransformRidgelet Tiling Divides the frequency domain to dyadic coronae |  |  [2 s, 2 s+1 ]. In the angular direction, samples the s -th corona at least 2 s times. In the radial direction, samples using local wavelets. Fourier Transform within Tiling

18 The Ridgelet Transform The ridgelet element has a formula in the frequency domain: where,  i,l are periodic wavelets for [- ,  ). i is the angular scale and l  [0, 2 i-1 –1] is the angular location.  j,k are Meyer wavelets for . j is the ridgelet scale and k is the ridgelet location.

19 Ridgelet Analysis Each normalized square is analyzed in the ridgelet system: The ridge fragment has an aspect ratio of 2 -2s  2 -s. After the renormalization, it has localized frequency in band |  |  [2 s, 2 s+1 ]. A ridge fragment needs only a very few ridgelet coefficients to represent it.

20 Digital Ridgelet Transform ( DRT ) Unfortunately, the (current) DRT is not truly orthonormal. An array of n  n elements cannot be fully reconstructed from n  n coefficients. The DRT uses n  2n coefficients for almost perfect reconstruction Still a lot of research need to be done…

21 Curvelet Transform The four stages of the Curvelet Transform were: Sub-band decomposition Smooth partitioning Renormalization Ridgelet analysis

22 Image Reconstruction The Inverse of the Curvelet Transform: Ridgelet Synthesis Renormalization Smooth Integration Sub-band Recomposition

23 Example: Roy Lichtenstein: “In The Car” 1963 Original Image (256  256) Approximation with only 64 wavelets and 256 curvelets (about 0.5% of the coefficients)

24 Original Adding Gaussian Noise WT + Thresholding WT + k-  Thresholding Curvelet Transform Example: Noise Reduction using Curvelet transform.

25 Original Adding Gaussian Noise WT + Thresholding WT + k-  Thresholding Curvelet Transform Example: Noise Reduction using Curvelet transform.

26 References [1]D.L. Donoho and M.R. Duncan. Digital Curvelet Transform: Strategy, Implementation and Experiments; Technical Report, Stanford University 1999 [2]E.J. Candès and D.L. Donoho. Curvelets – A Surprisingly Effective Non-adaptive Representation for Objects with Edges; Curve and Surface Fitting: Saint Malo 1999 [3]Lenna examples from


Download ppt "Introduction to the Curvelet Transform By Zvi Devir and Yanai Shpinner."

Similar presentations


Ads by Google