18cse390t U1 s9 Slo1 2 Content
Uploaded by
Deepa S18cse390t U1 s9 Slo1 2 Content
Uploaded by
Deepa S18CSE390T
Computer Vision
U1 - S9 - SLO1&2 - Two-Dimensional Fourier Transforms, Wiener
Filtering
Two-Dimensional Fourier Transforms
The Fourier transform is simply a tabulation of the magnitude and phase response at
each frequency
Continuous domain:
Discrete domain:
Two-Dimensional Fourier Transforms
All of the Fourier transform properties from 1D carry over to two dimensions if we
replace the scalar variables x, , x0 and a, with their 2D vector counterparts
• x = (x, y)
• = (x, y)
• x0= (x0, y0)
• a = (ax, ay)
Wiener Filtering
• Unlike direct Inverse Filtering, which is poor in general, we want to handle noise
more explicitly
• The objective is to find an estimate, of the uncorrupted image, such that the
mean square error between them is minimized
• This error is given by
• where E{.} is the expected value
• We assume that the noise (or image) has zero mean, the noise and an image are
uncorrelated, and the gray levels in the estimate are linear functions of the level
in the degraded image
Wiener Filtering
• To minimize
• Where H(u,v) is the degradation function, H*(u,v) its complex conjugate,
• |H(u,v)|2=H(u,v)H*(u,v) the power spectrum, and Sn, Sf are the power spectra of
the noise and the undegraded image respectively
Wiener Filtering
• Applying the approximate Wiener Filter
Wiener Filtering
• Applying the approximate Wiener
Filter, K chosen to get the best visual
result
Wiener Filtering
• One problem we have had so far in both the inverse and Wiener methods is the
need to know about the degradation function H(u,v).
• With the Wiener filter, need to know the power spectra of un-degraded image
and noise to get optimum results (although we used K in the approximate form).
• Constrained Least Squares Filtering gives an optimum restoration for each image
to which it is applied
