Notes 20110208 CIS 6320 Image Processing - Presentation 1 -- Zaman A N K
From SnOwy - Ed's Wiki Notebook
Presentation 1 -- Zaman A N K
- submit evaluation given -- http://www.socs.uoguelph.ca/~matsakis/CIS6320/sched.htm
Contents |
Applications of Fourier Transforms
- a sum of period functions
- periodic functions can be expanded into sine and cosine functions
- frequencies in images
- high frequencies -- quickly varying information e.g. edges
- low frequencies -- slowly varying information e.g. continuous surface
- discrete Fourier transform (DFT) -- a certain form of the transform
- frequency domain image filtering
- f(x, y) input image → g(x, y) enhanced image
- pre-processing f(x, y) → Fourier transform → processing → inverse Fourier transform → g(x, y)
- frequency domain tranformation
Image Enhancement
- lowpass filters -- Ideal Lowpas Filter (ILPF)
- H(u, v) = {1 if D(u,v) -- I'll have to look this up
- everything is with respect to the middle point of the image
- sharp discontinuity
- http://en.wikipedia.org/wiki/Low-pass_filter#Ideal_and_real_filters ?
- Butterworth Lowpass Filter
- smooth transfer function, no sharp discontinuity
- Gaussian Lowpass Filter (GLPF)
- smoother than previous two lowpass filters
- examples of lowpass filters
- images can be blurred by attenuating high frequency ...
- broken characters were blurred on this screen where g(x, y) showed interpolation?
- Ideal Highpass Filter (IPHF)
- Problems with ringing -- as we increase D0 -- allows us to reduce ringing effect
- Butterworth Highpass Filter (BHPF)
- smoother results
- Gaussian Highpass Filter
- note: look up the transfer functions for Guassian high/low filters
- Gaussian performs in between
Differences between highpass and lowpass filters
- allows different bands (look up) of frequencies through
- transfer function is step / butterworth / guassian
Convolution
- relates frequency and spatial domains
Image Compression
- lossy and lossless
- trade off between quality and compression ratio
- want to keep lossless as much as possible (Huffman)
- lossy e.g. JPEG
Data Vs Information
- data: raw facts
- information: what we get from processing data
- different amount of data can provide the same amount of information
Data redundancy
- coding / psychovisual / interpixel
- reduce these things
- avoid redundancy: codes selected according to probabilities of events
- Variable Length Coding
- psychovisual redundancy
- peculiarities of the human visual ystem
- visual system does not respond equally to all visual stimulus
- interpixel redundancy
- any pixel value can be reasonably predicted by its neighbours (correlated)
- reduce interpixel redundancy -- transformed into different format (transformation)
- thresholding -- differences between adjacent pixels, DFT.
Image Analysis
later
Image Reconstruction
later
