02 - Image Enhancement

Question 1

Influence image quality

Answer 1

• resolution
• noise
• contrast
• inhomogeneous contrast
• artifacts

Question 2

Noise types

Answer 2

• types
• – spatially correlated
• – event
• – structured noise
• model
• – Gaussian (additive)
• – Poisson (multiplicative)
• – salt’n’pepper (outlier noise)
• SNR = avg(x)/std(x)

Question 3

Filtering

Answer 3

• def = processing unit that enhances the wanted info and suppresses the unwanted info
• LP, HP filters
• linear, LP = box, Gauss
• median filter (median value of surrounding pixels)
• linear, HP = gradients, Laplacian, Sobel
• frequency-space filters (Fourier transform)

Question 4

Scale spaces

Answer 4

• filter on different scales
• wavelets: decompose signal in 2 scales -> noise reduction
• parametrized scales space -> diffusion stops at edges (heat transport; non-linear diffusion; …)

Question 5

Non local means

Answer 5

• smoothing with avgg similar intensities in a global sense
• weigh one pixel with avg of all others
• good smoothing effect with strong thin lines
• complexity N^2

Question 6

Validation image enhancement

Answer 6

• test with phantom img and relevant features
• labeled data
• metrics = mean square error, structural similarity index

Question 7

How to suppress noise and artifacts?

Answer 7

convolution, median filters, wavelet denoising, PDE filters