Frequency Domain Methods

Frequency Domain Methods Major filter categories Example

f(x,y) |F(u,v)|

Spatial Domain Frequency Domain Filters are classified based on their properties in the Original signal
frequency domain:
H(u,v)
Low-pass filtered
element-wise (1) Low-pass
complex
multiplication
G(u,v)
(2) High-pass High-pass filtered

H(u,v) is specified directly (3) homomorphic filter

The time-domain (or spatial-domain) signals have both continuous and infinite domains. There
is no explicit or implied periodicity in it. This is where we use the Fourier transform, which is in the frequency domain.
a way of representing images in discrete form. We use the Fourier transform to represent Band-pass filtered
images in the frequency domain.
g(x,y)
• The frequency domain (FD) method converts the signal from the time domain to the
frequency domain by a fast Fourier transform (FFT), Band-stop filtered

June 23 DR JOSHUA ALACKAL 2 June 23 DR JOSHUA ALACKAL 3 June 23 DR JOSHUA ALACKAL 4 June 23 DR JOSHUA ALACKAL 5

Low-pass (LP) filtering Low-pass filtering (cont’d) Lowpass filtering (cont’d) Lowpass (LP) filtering (cont’d)
In 2D, the cutoff frequency D0 is specified by the radius of a circle Assuming that we have centered the filter at (P/2, Q/2), as required, D(u,v)
Preserve low frequencies. Preserves low frequencies, attenuates high frequencies. centered at the origin. should be computed as:
◦ Useful for removing details and noise suppression.
frequency domain 1D ILPF 2D ILPF 2D ILPF
representative 1 if 0  u  D0 1 if D (u , v )  D0
example H (u ) =  1 if D(u, v)  D0 H (u , v) = 
0 otherwise H (u, v) =  0 otherwise
0 otherwise
Example: where D(u , v) = (u − P / 2) 2 + (v − Q / 2) 2
where D(u, v) = u + v 2 2

D0: cut-off frequency – user specified

June 23 DR JOSHUA ALACKAL 6 June 23 DR JOSHUA ALACKAL 7 June 23 DR JOSHUA ALACKAL 8 June 23 DR JOSHUA ALACKAL 9

ILPF performance
High-pass (HP) filters High-Pass filtering (cont’d) High-Pass filtering (cont’d)
freq. domain spatial domain Preserves high frequencies, attenuates low frequencies.
Preserve high frequencies.
• A high-pass filter can be obtained from a low-pass filter:
◦ Useful for highlighting details.
ILPF decreasing the 1D IHPF 2D IHPF
disparity between pixel
values by averaging nearby H HP (u , v) = 1 − H LP (u , v)
frequency
pixels. LPF tends to retain representative
domain
the low frequency example
sinc information within an image
Example:
1 if D(u, v)  D0
H (u, v) = 
1 if u  D0 0 otherwise
H (u ) =  =1-
* = 0 otherwise where D(u, v) = u 2 + v 2
or D(u, v) = (u − P / 2) 2 + (v − Q / 2) 2
if centered
June 23 DR JOSHUA ALACKAL 10 June 23 DR JOSHUA ALACKAL 11 June 23 DR JOSHUA ALACKAL 12 June 23 DR JOSHUA ALACKAL 13

Butterworth high pass filter (BHPF)

Butterworth LP filter (BLPF) Butterworth LP filter (cont’d)
BHPF is used for image sharpening in the frequency domain. Image Sharpening is
Butterworth Lowpass Filter (BLPF) Assuming that we have centered the filter at (P/2, Q/2), as a technique to enhance the fine details and highlight the edges in a digital image.
is used for image smoothing in required, D(u,v) should be computed as: It removes low-frequency components from an image and preserves high-
the frequency domain. It removes frequency components.
high-frequency noise from a
digital image and preserves low- BLPF BHPF
1 1
frequency components 1- H (u , v) =
1 + [ D(u , v) / D0 ]2 n
H (u , v) =
1 + [ D0 / D(u, v)]2 n
1
H (u, v) = Approaches an ILPF as n increases!
1 + [ D(u, v) / D0 ]2 n n: order of filter
n=1 n=4 n=16
where D(u, v) = u 2 + v 2 1
H (u , v) =
1 + [ D(u , v) / D0 ]2 n
n: order of filter
where D(u, v) = (u − P / 2) 2 + (v − Q / 2) 2

June 23 DR JOSHUA ALACKAL 14 June 23 DR JOSHUA ALACKAL 15 June 23 DR JOSHUA ALACKAL 16