Chp.4 - Neighbourhood Operators Flashcards
The two main categories and examples (2 each)
Linear: convolution operators, spatial averaging
Non-linear: median filtering, morphological operators
The physical meaning of the equation (linear neighbourhood operator)
g(m, n) = summation summation (m’, n’∈R) f(m’, n’)h(m − m’, n − n’)
Place the flipped mask at every possible location in the image and at the location in the image, compute an inner product between the mask and the local image region.
AKA linear, spatially invariant filtering.
General representation for local, linear operators is:
g(m, n) = summation summation (m’, n’∈R) f(m’, n’)h(m − m’, n − n’)
where h: convolution mask
R: local region for summation, determined by mask size
Describe the process for smoothing
1) generate values of
Edge enhancement is which type of filtering?
high pass filter
Reasons/types of blurring and how to solve them respectively
1) due to “defocus”, isotropic sharpening is required
2) cause by motion, anisotropic filtering is better
What are filter banks?
Sets of separable or non-separable filters (in the form of convolution masks) which are all applied to the same image in parallel.
Usage of KL transform
satellite images and ideal data compression but not practical
Applications of KL transform
satellite images and ideal data compression but not practical
Applications of Hadamard transform
texture synthesis and tracking algorithms of pathogens in MRI
compression theory
reducing the storage spaces by only taking the important transform coefficients
typically: cosine discrete transform, wavelet transform works too
edge enhancement theory
suppress noise will depend on the type of noise
noise: localised/distributed across discrete Fourier space
image structure: concentrated in the region of Fourier that corresponds to slowly varying frequency/component
take the image into transform space and then apply a filtering operation
CT scan working principle
1) Take 1D images across the patient
2) Apply FT to each of the image projection
3) Take each of the FT into a 2D Fourier space
(like different lines across a circular 2D Fourier space)
4) Apply 2D inverse DFT to reconstruct the image
How’s MRI different with CT scan in principle and why?
MRI required 2D and 3D Fourier space as MRI relies on frequency and phase coding.
Gaussian scale space defintion and why is it useful
A series of progressively smoothed images with Gaussian mask
It is useful for some type of tasks such as image registration as the possibility to be stuck at the local minima is reduced in a cruder image than a finer image
