Image Processing 1 + 2 Flashcards
Digitization
Limited x resolution (Sampling) and Limited y resolution (Quantization)
Spatial Quality
Pixel size
Intensity Quality
Grey scale bit level
Nyquist-Shannon Theorem
Min. sampling rate = 2f, lower means details are lost (like clothing on tv) and new patterns can arise
Convolution/Kernel
CONVOLUTION is the process of applying a KERNEL to the pixels of an image
Dirac/Delta Kernel
Kernel width approaches 0 (and height infinity)
Mach Band Effect
How the brain enhances contrast
Aliasing
High frequency patterns become (different) low frequency patterns
Point Spread Function (PSF)
Determines the optical resolution, the smallest spatial detail visible
Gamma correction
0-1 = lighter 1+ = darker
More contrast (l)
- Contrast stretching (slope streched)
- Windowing (area with slope narrowed _/-)
- Thresholding (instant slope _|-)
- Double thresholding/window slicing (|-|)
Histogram Equalization
Remap grey values so they all occur in equal numbers (can only be approached for discrete, digital images)
Ideal pixels per bin
Npixels/Ngreyscales
Cumulative histogram
Number of pixels with at MOST a grey value x
Grey scale substraction/addition
Shows change/filters out noise after dividing by amount of images added
2D Histogram
The grey values in the 2D histogram indicates how often certain combinations of grey values occur
Noise Filters
- Salt and Pepper (adds some black and white)
- Uniform (equally around the range of the original)
- Gaussian
Convolution/Correlation
Cross-multiplication/normal multiplication
Convolution has laws, correlation doesn’t:
-Cummutative law (order filter/image irrelevant)
-Associative law (order of multiple filters irrelevant)
-Distributive law gf+hf=(g+h)f
Filter near image border
-Fixed values
-Extrapolation
Mirroring
Low pass/High pass filtering
Detail enhancement/Unsharp masking
Low pass: Reduces noise
High pass: image - low pass, (sums up to 0 instead of 1 ,so constant areas vanish)
Detail enhancement: high pass, but adds up to 1 (constant areas stay the same, contrast enhanced)
Unsharp masking: + a high pass filtered image or - a low pass filtered image
