Image Coding, Pixel Operations,Spatial FIltering and Images In Frequency Flashcards
What is sampling & quantization
Sampling is Digitizing in coordinates. From continuous signal to discrete captured by finite samples.
𝑓 𝑚, 𝑛 ≜ 𝑓(𝑚Δ𝑥, 𝑛Δ𝑦)
Quantization is digitizing in amplitude! = 𝑓𝑑 𝑚, 𝑛 = 𝑄[𝑓 𝑚, 𝑛 ]
Spatial & Grey Level resolution & Contrast
Spatial Resolution: Smallest detectable detail. Nr of pixels needed per unit distance
Grey Level Resolution is smallest detectable change in grey level. Nr of bits per Pixel
Contrast is the difference between highest and lowest intensity level.
Types of Transformations
1- Geometric (scaling rotating ..)
2- Single Pixel Operations(intensity transform,histogram eq, output depends on init values)
3-Local Operations (Filters + neigbors)
Geometric transform
Modifies spatial relationship among pixels.
(x,y) = T{(x,y)}
Works on geometrical points
Types of geometric
Translation
Euclidean (rotates)
Similarity (zvogelohet)
Affine (paralelogram)
prjective (ruan vec vije te drejta)
AffineTransformation
Linear
Perserves
1-Collinearity
2-Ratios on a line
Forward and backward mapping
Mapping : From Geometric points (coordinate transformations) Back to pixels
1)Forward mapping
a)Compute corresponding pixel
b)Ambiguity:many points correspond to the same one
c)Resampling needed
d)ambiguity-> missing pixels
2)Backward mapping
a)Foreach next pixel, compute init pixel
b)Finds 1 corresponding point foreach
c)needs resampling
d)fills all pixels
e)better
Single Pixel Operations
Change grey level on image.
I(x,y) is image
T(.) is grey level change
usage:Enhance Contrast, threshold
Types of SPO
1-Negative(switch dark and light)
2- Log (highlight differences among pixels)
3-Gamma (like log but tunable)
4-COntrast Stretching and thresholding
5-Slicing
6-Histogram Equalization
Usage:
a)Image Statistics
b)Compression
c)Segmentation
d)Image enhancment
Invertable if strictly monotonic.
Flatens histogram
Types of SPO
1-Negative(switch dark and light)
2- Log (highlight differences among pixels)
3-Gamma (like log but tunable)
Local Equalization & Histogram Specification
Useful when different regions have different pixel distribution.
Histogram Specification:
a) Equalize histogram
b) Specify shape.
c) Obtain Inverse Transform
d) Apply Equalization and then transformation to desired shape
e)Map Together
Equalization -> Bad if nr of pixels with low gray level is high
Specification Better,
Local Operations
Defined on a kernel=neighborhood of pixels, each one has a weight
Performed on spatial domain
Types = Correlation (Linear + Nonlinear)
Min Max etc
Correlation
Filter superimpsed on each pixel
Evaluation of a weighted average
To not change image brightness, image is normalized where sum of all weights is 1.
Linear Filters Average
a) Averaging (smoothing and restoration)
-large filter, stronger smoothing
-nxn filter can be seperated in nx1 and 1xn filters-> faster (OMNa+b)
-Benefits-> Noise reduction + blur/smoothing
Linear Filters Derivative
b)Derivative(Sharpening)
1-st order: Gradient,Gradient Module, Approx Gradmint module/
Linear +Isotropic, NonLinear Nonlinear
2nd order: Laplacian + approximation that are Linear And isotropic
Lapacioan -> Enhance transition
Laplacian + original -> sharpened version of original.
