Documente Academic
Documente Profesional
Documente Cultură
Edges (derivatives):
• Gaussian pyramids
• Laplacian Pyramids f(x)
• Wavelet Pyramids
• Applications
f (x)
Image Pyramids
Image pyramids
Image Pyramid = Hierarchical representation of an image
• Gaussian Pyramids
• Laplacian Pyramids
No details in image -
Low
Resolution (blurred image) • Wavelet/QMF
low frequencies
*
=
*
~
=
=
*
=
Image Pyramid
Gaussian Pyramid
Low resolution
Level n
1X1
Level 1
2n-1 X 2n-1
Level 0
2n X 2n
High resolution
Normalized: Σwi = 1
Symmetry: wi = w-i
w-2 w-1 w0 w1 w2
w-2 w-1 w0 w1 w2
Gaussian Pyramid Gaussian Pyramid
Equal Contribution: for all j Σwj+2i = constant Equal Contribution: for all j Σwj+2i = constant
w0
w-2 w-2
Gaussian Pyramid
For a = 0.4 most similar to a Gauusian filter
Burt & Adelson (1981)
g = [0.05 0.25 0.4 0.25 0.05]
low_pass_filter = g’ * g =
a + 2b + 2c = 1 0.2
a + 2c = 2b
0.15
0.1
a > 0.25
b = 0.25 0.05
c = 0.25 - a/2
0
5
4 5
3 4
3
2
2
1 1
Gaussian Pyramid - MultiScale Pattern Matching
Computational Aspects
Option 1:
Scale target and search for each in image.
Memory:
2NX2N (1 + 1/4 + 1/16 + ... ) = 2NX2N * 4/3
Computation:
i times
Option 2:
Example:
Search for original target in image pyramid.
h2 = * =
g g
search
search
search
Laplacian Pyramid
Image pyramids Motivation = Compression, redundancy removal.
compression rates are higher for predictable values.
e.g. values around 0.
• Gaussian Pyramids
G0, G1, .... = the levels of a Gaussian Pyramid.
• Laplacian Pyramids
• Wavelet/QMF Predict level Gl from level Gl+1 by Expanding Gl+1 to G’l
Gl+1
Expand
Reduce
Gl G’l
Ll = Gl - G’l
What does blurring take away? What does blurring take away?
Gaussian Laplacian
Pyramid Pyramid
expa
nd
ex -
pa
=
nd
ex -
pa
=
nd
- =
smoothed – original
Laplacian Gl = Ll + G’l
Pyramid
expand
+ =
expand
+ =
expand
+ =
Original
= Image
from: B.Freeman
Memory:
2NX2N (1 + 1/4 + 1/16 + ... ) = 2NX2N * 4/3
However coefficients are highly compressible.
Computation:
- =
Registration
hi-1 hi
ki
k1 k2 k3
Image Blending Blending
High Frequencies
High Frequencies
Low Frequencies
Low Frequencies
Multiresolution Spline - Example Multiresolustion Spline - Using Laplacian Pyramid
Wide Transition
laplacian level 4
laplacian level 2
laplacian level 0
© prof. dmartin
What is a good representation
Image pyramids
for image analysis?
• Gaussian Pyramids • Pixel domain representation tells you
• Laplacian Pyramids “where” (pixel location), but not
“what”.
• Wavelet/QMF – In space, this representation is too localized
• Fourier transform domain tells you
“what” (textural properties), but not
“where”.
– In space, this representation is too spread out.
• Want an image representation that gives
you a local description of image
events—what is happening where.
– That representation might be “just right”.
Spatial Spatial
Freq. Freq.
Fourier basis Fourier basis
Spatial Spatial
Freq. Freq.
Wavelet basis Wavelet basis
Spatial Spatial
Various Wavelet basis Wavelet - Frequency domain
Wavelet bands are split recursively
image
H L
H L
H L
Frequency domain
Fourier Space
Wavelet Transform - Example Wavelet Transform - Example
Clean image
Noisy image
Mappingfunctions
Mapping functions
Transform
Transform xiw
I W
W
yiw
Inverse
Inverse
Transform
Transform
WTT
W
I clean
Image Pyramids - Comparison
Filters:
Gaussian Pyramid
Laplacian Pyramid
Wavelet Pyramid
Localized in space r r
F = Uf
Delta Standard Not localized in Frequency
From: B. Freeman
* = = *
= = *
*
Pyramid Level 2
Pyramid Level 2
= * *
From: B. Freeman
Pyramid Level 2 0 0 1 4 6 4 1 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 1 4 6 4 1 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 1 4 6 4 1 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 1 4 6 4 1 0 0 0 0 0 0 0
= * * 0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
0
4
0
6
1
4
4
1
6
0
4
0
1
0
0
0
0
0
0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 4 6 4 1 0
from: B.Freeman
Gaussian Pyramid Laplacian Pyramid
= = * = *
pixel image pixel image
Wavelet Transform
= *
The End
Wavelet
Ortho-normal pixel image
pyramid
transform (like
Fourier transform),
but with localized
basis functions.
From: B. Freeman