1
Q
Medical Imaging
A
- X-Ray
- CT
- MRI
- Ultrasound
2
Q
Nyquist Sampling Theory
A
- in order to reach perfect reconstruction of input signal which has f, sampling must occur at 2f
- phase still issue
3
Q
Image Quality Contrast
A
- Global contrast: (Imax-Imin/imax+Imin), good for detection of inefficient use of available intensity range, neglegts distribution of intensities
- Root Mean Square Contrast: no differeation between distributions with different detail levels if dist is same
- Gray-Level Co-occurence: measure weights the cooccurences of intensities by their difference, strong differences at edges result in higher contrast value
4
Q
Noise
A
- unwanted image-corrupting influence
- random fluctuation of intensities with zero mean
- object detection depends on ratio of object-background contrast to noise variance
5
Q
Signal to Noise Ratio
A
- ratio of meaningful signal information and unwanted signal
- average signal value / standard deviation of the background
6
Q
Image Quality - Edges
A
- fist and second spacial derivation of image is sensitive to edges
7
Q
Image Enhancement - lookup table
A
- transfer function to define windows for parts of the original intensity range
8
Q
Image Enhancement - Histogram equalization
A
- create new image with constant histogram
- all intensities cinsidered equally important
- Centropy stary the same
9
Q
Edge Enhancement
A
- gradient calculation (approximation via differences)
- Gaussian Filters: Laplacian of Gaussians
- Second derivatives: zero crossing is edge (calculated wit LoG)
10
Q
Feature Enhancement - Vesselness
A
- Hessian: 2nd derivatives
- eigenval1, very small
- eigenval2,3, large and nearly equal
11
Q
Noise reduction - linear filtering
A
- I locally constant, E is avg of local neighborhood
- the bigger the filter the blurrier
- problem: ringing, solution butterworth filter (attenuates noise proportional to frequenzy with cut off)
12
Q
Noise reduction - median filtering
A
- selection of E from ordered list of values in neighborhood
- edges are preserved if: edge is traight within neighborhood region, signal diff exceeds noise amplidute
13
Q
Noise reduction - diffusion filtering
A
- smoothing at edges without changing them
- image intensity considered material density
- noise considered density variation
- homogeneous / inhomogenous diffusion
14
Q
Noise reduction - Bayesian Image Restoration
A
- representation of image characteristics by markov random field
- search for image I that maximizes cond probability of observing noisy image In
15
Q
image enhancement facts
A
- extract information or suppress artefacts
- compromise: trades generality for accuracy
- edge preserving smoothing con lead to false boundaries
