EECE253 08 FrequencyFiltering

EECE\CS 253 Image Processing
Lecture Notes: Frequency Filtering Lecture Notes
Richard Alan Peters II

Department of Electrical Engineering and Computer Science
Fall Semester 2007
This work is licensed under the Creative Commons Attribution-Noncommercial 2.5 License. To view a copy of this license, visit http://creativecommons.org/licenses/by-nc/2.5/ or send a letter to Creative Commons, 543 Howard Street, 5th Floor, San Francisco, California, 94105, USA.
Convolution Property of the Fourier Transform

Let functions f ( r, c ) and g ( r, c ) have Fourier Transforms F (u, v ) and G (u, v ). Then,
* = convolution = multiplication
F{ f g } = F G.
Moreover,
F{ f g } = F G.
The Fourier Transform of a convolution equals the product of the Fourier Transforms. Similarly, the Fourier Transform of a convolution is the product of the Fourier Transforms
August 11, 2012
1999-2007 by Richard Alan Peters II
Convolution via Fourier Transform
Image & Mask
Transforms
Pixel-wise Product
Inverse Transform
August 11, 2012
How to Convolve via FT in Matlab

1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. Read the image from a file into a variable, say I. Read in or create the convolution mask, h. The mask is usually 1-band Compute the sum of the mask: s = sum(sum(h)); If s == 0, set s = 1; For color images you may need to do each step for Create: H = zeros(size(I)); each band separately. Copy h into the middle of H. Shift H into position: H = ifftshift(H); Take the 2D FT of I and H: FI=fft2(I); FH=fft2(H); Pointwise multiply the FTs: FJ=FI.*FH; Compute the inverse FT: J = real(ifft2(FJ)); Normalize the result: J = J/s;
August 11, 2012
Coordinate Origin of the FFT

Even Odd Even
Center = (floor(R/2)+1, floor(C/2)+1)

Odd
Image Origin
Image Origin
Weight Matrix Origin
Weight Matrix Origin
After FFT shift
After FFT shift
After IFFT shift
After IFFT shift
August 11, 2012
Matlabs fftshift and ifftshift

J = fftshift(I): I (1,1) J ( R/2 +1, C/2 +1)
5 6 4 8 9 7 1 2 3
7 5 2 3 1 2 3 1 8 9 6
9 4
I = ifftshift(J): J ( R/2 +1, C/2 +1) I (1,1) where x

August 11, 2012
= floor(x) = the largest integer smaller than x.

2 3 1
Blurring: Averaging / Lowpass Filtering

Blurring results from:
q
Pixel averaging in the spatial domain:

Each pixel in the output is a weighted average of its neighbors. Is a convolution whose weight matrix sums to 1. High frequencies are diminished or eliminated Individual frequency components are multiplied by a nonincreasing function of such as 1/ = 1/(u2+v2). The values of the output image are all non-negative.
Lowpass filtering in the frequency domain:

August 11, 2012
Sharpening: Differencing / Highpass Filtering

Sharpening results from adding to the image, a copy of itself that has been:
q
Pixel-differenced in the spatial domain:

Each pixel in the output is a difference between itself and a weighted average of its neighbors. Is a convolution whose weight matrix sums to 0. High frequencies are enhanced or amplified. Individual frequency components are multiplied by an increasing function of such as = (u2+v2), where is a constant. The values of the output image positive & negative.
Highpass filtered in the frequency domain:

August 11, 2012
Recall:
Convolution Property of the Fourier Transform

Let functions f ( r, c ) and g ( r, c ) have Fourier Transforms F (u, v ) and G (u, v ). Then,
* = convolution = multiplication
F { f g } = F G.
Moreover,
F { f g } = F G.
Thus we can compute f g by
{ f g = F -1 F G} .
August 11, 2012
The Fourier Transform of a convolution equals the product of the Fourier Transforms. Similarly, the Fourier Transform of a convolution is the product of the Fourier Transforms
Ideal Lowpass Filter

Multiply by this, or convolve by this
Image size: 512x512 FD filter radius: 16
Fourier Domain Rep.
Spatial Representation
Central Profile
August 11, 2012
10

Fourier Domain Rep.
Central Profile
August 11, 2012
11
Consider the image below:
Power Spectrum and Phase of an Image
Original Image
Power Spectrum
Phase
August 11, 2012
12
Original Image
Power Spectrum
Ideal LPF in FD
August 11, 2012
13
Filtered Image
Filtered Power Spectrum
Original Image
August 11, 2012
14
Ideal Highpass Filter

Image size: 512x512 FD notch radius: 16
Fourier Domain Rep.
Central Profile
August 11, 2012
15
Original Image
Power Spectrum
Ideal HPF in FD
August 11, 2012
16
signed image; 0 mapped to 128
Filtered Image*
Original Image
August 11, 2012
17
Positive Pixels
Filtered Image*
Negative Pixels
August 11, 2012
18
The Uncertainty Relation

space frequency
FT
space
frequency
FT
A small object in space has a large frequency extent and vice-versa.

19
August 11, 2012
The Uncertainty Relation

frequency small extent space
IFT
Recall: a symmetric pair of impulses in the frequency domain becomes a sinusoid in the spatial domain. A symmetric pair of lines in the frequency domain becomes a sinusoidal line in the spatial domain.
large extent
small extent frequency large extent
large extent space
small extent
IFT
small extent
large extent 1999-2007 by Richard Alan Peters II
August 11, 2012
20
Ideal Filters Do Not Produce Ideal Results
IFT
A sharp cutoff in the frequency domain
causes ringing in the spatial domain.
August 11, 2012
21
Ideal Filters Do Not Produce Ideal Results
Ideal LPF
Blurring the image above w/ an ideal lowpass filter

August 11, 2012
distorts the results with ringing or ghosting.

22
Optimal Filter: The Gaussian
IFT
The Gaussian filter optimizes the uncertainty relation. It provides the sharpest cutoff with the least ringing.
August 11, 2012
23
One-Dimensional Gaussian
g ( x) =
August 11, 2012
1 1
( x )1 1 1
24
Two-Dimensional Gaussian
R = 512, C = 512
If and are different for r & c
or if and are the same for r & c.
= 257, = 64
August 11, 2012
25
Optimal Filter: The Gaussian
Gaussian LPF
With a gaussian lowpass filter, the image above

August 11, 2012
is blurred without ringing or ghosting.

26
Gaussian Lowpass Filter

Image size: 512x512 SD filter sigma = 8
Fourier Domain Rep.
Central Profile
August 11, 2012
27

Fourier Domain Rep.
Central Profile
August 11, 2012
28
Original Image
Power Spectrum
Gaussian LPF in FD
August 11, 2012
29
Filtered Image
Original Image
August 11, 2012
30
Gaussian Highpass Filter

Image size: 512x512 FD notch sigma = 8
Fourier Domain Rep.
Central Profile
August 11, 2012
31
Original Image
Power Spectrum
Gaussian HPF in FD
August 11, 2012
32
Filtered Image*
Original Image
August 11, 2012
33
Positive Pixels Filtered Image
Filtered Image*
Negative Pixels
August 11, 2012
34
Comparison of Ideal and Gaussian Filters
Ideal LPF
Original Image
Ideal HPF*
August 11, 2012
35
Gaussian LPF
Original Image
Gaussian HPF*
August 11, 2012
36
Another Highpass Filter
original image original image
filter power spectrum filter power spectrum
filtered image* filtered image*
August 11, 2012
37
Ideal Bandpass Filter
filter power spectrum filter power spectrum
filtered image* filtered image*
August 11, 2012
38
Gaussian Bandpass Filter
Image size: 512x512 sigma = 2 - sigma = 8
Fourier Domain Rep.
Central Profile
August 11, 2012
39
Original Image
Power Spectrum
Gaussian BPF in FD
August 11, 2012
40
Filtered Image*
Original Image
August 11, 2012
41
Positive Pixels Filtered Image
Filtered Image*
Negative Pixels
August 11, 2012
42
Ideal BPF*
Original Image
Gaussian BPF*
August 11, 2012
43
Power Spectrum and Phase of a Blurred Image
blurred image blurred image

August 11, 2012
power spectrum power spectrum

phase phase
44
Power Spectrum and Phase of an Image

August 11, 2012

phase phase
45
Power Spectrum and Phase of a Sharpened Image
sharpened image sharpened image

August 11, 2012

phase phase
46

EECE253 08 FrequencyFiltering

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

EECE253 08 FrequencyFiltering

Hochgeladen von

Copyright:

Verfügbare Formate

EECE\CS 253 Image Processing

Lecture Notes: Frequency Filtering Lecture Notes

Richard Alan Peters II

Convolution Property of the Fourier Transform

August 11, 2012

1999-2007 by Richard Alan Peters II

Convolution via Fourier Transform

Image & Mask

August 11, 2012

1999-2007 by Richard Alan Peters II

How to Convolve via FT in Matlab

August 11, 2012

Coordinate Origin of the FFT

Center = (floor(R/2)+1, floor(C/2)+1)

Weight Matrix Origin

Weight Matrix Origin

After FFT shift

After FFT shift

After IFFT shift

After IFFT shift

August 11, 2012

1999-2007 by Richard Alan Peters II

Matlabs fftshift and ifftshift

I = ifftshift(J): J ( R/2 +1, C/2 +1) I (1,1) where x

= floor(x) = the largest integer smaller than x.

1999-2007 by Richard Alan Peters II

Blurring: Averaging / Lowpass Filtering

Pixel averaging in the spatial domain:

Lowpass filtering in the frequency domain:

August 11, 2012

1999-2007 by Richard Alan Peters II

Sharpening: Differencing / Highpass Filtering

Pixel-differenced in the spatial domain:

Highpass filtered in the frequency domain:

August 11, 2012

1999-2007 by Richard Alan Peters II

Convolution Property of the Fourier Transform

1999-2007 by Richard Alan Peters II

Ideal Lowpass Filter

Image size: 512x512 FD filter radius: 16

Fourier Domain Rep.

August 11, 2012

1999-2007 by Richard Alan Peters II

Ideal Lowpass Filter

Image size: 512x512 FD filter radius: 8

Fourier Domain Rep.

August 11, 2012

1999-2007 by Richard Alan Peters II

Consider the image below:

Power Spectrum and Phase of an Image

August 11, 2012

1999-2007 by Richard Alan Peters II

Ideal Lowpass Filter

Image size: 512x512 FD filter radius: 16

August 11, 2012

1999-2007 by Richard Alan Peters II

Ideal Lowpass Filter

Image size: 512x512 FD filter radius: 16

Filtered Power Spectrum

August 11, 2012

1999-2007 by Richard Alan Peters II

Ideal Highpass Filter