Subba d eco pos t o using te a s Subband Decomposition us g Filter Banks
x(n1,n2)
Analysis section Filter bank (Analysis)
: :
Processing section
: :
Processing section
: :
Synthesis section Filter bank (Synthesis)
xr(n1,n2)
Input image Subband images
The
Qua t at o Quantization
Quantization: main component in source coders
Problem: Original image is often very large Reduce the data Remove redundancy y no loss
info content preserved p
Prediction (DPCM, ADPCM) Entropy coding Transform coding (e.g., trans
Subba d eco pos t o using te a s Subband Decomposition us g Filter Banks
x(n1,n2)
Analysis section Filter bank (Analysis)
: :
Processing section
: :
Processing section
: :
Synthesis section Filter bank (Synthesis)
xr(n1,n2)
Input image Subband images
The
Transforms a so s
Basic tools in image processing g p g Represent images by a series of coefficients which
can be used for processing and analysis General form for N1xN2 image
X K1 , K 2
N1 1 N 2 1 n1 0 n2 0
x n1 , n2
n1n2
K1 , K 2 ;
Transform Kernel (Im
Optimal Transform: The Karhunen-Loeve Transform (KLT) p ( )
Recall: We are interested in unitary transforms because of their nice properties: energy conservation, energy compaction, decorrelation
Motivation: unitary
^ x
1
T
x (1D Transform; assume separab
Transforms a so s
Transforms that are commonly used are separable y p
Examples: Two-dimensional DFT, DCT, DST, Hadamard We can then use 1-D transforms in computing the 2D separable transforms:
Take 1-D transform of the rows => Xrows(K1,K2) Take 1-D trans
Vector qua t at o ecto quantization
Scalar quantizers are special cases of vector quantizers (VQ):
they are constrained to look at one sample at a time (memoryless) VQ does not have such constraint better RD perfomance expected
Source coding theorems and
Vector qua t at o ecto quantization
Example: Binary Splitting, M=8
centroid of the entire training data perturb centroid cluster w.r.t. new centroids compute new cluster centroids Advantage: Tree search can be used to reduce search complexity and computa
Wavelets a e ets
Wavelets: (signal processing interpretation)
A wavelet can be viewed as a bandpass filter with some upper and lower cutoff frequencies.
(t): "mother wavelet" or "a wavelet generating function"
used to generate a set of bandpass filters t
ARIZONA STATE UNIVERSITY
School of Electrical, Computer, and Energy Engineering EEE 508 Digital Image and Video Processing and Compression Spring 2012 Course Information Online using Blackboard under my.asu.edu http:/lina.faculty.asu.edu/eee508/
Class Hou
Qua t at o Quantization
Some sort of quantization is necessary to represent continuous
signals in digital form
x(n1,n2) x(t1,t2) t 2D Sampler Quantizer xq(n1,n2) n
Digitizer (A/D)
Quantization is also used for data reduction in virtually all lossy
codin
Visual sensation
Complex process, can be divided into 4 main stages:
Image formation
Incoming light is transformed by the optics of the eye and focused on the retina to create the retinal image. This is a series of simple and well known optical transfor
Wiener filter e e te
Non-zero mean signal and noise
In our derivation of the optimal LS solution, we assumed that signal and noise are zero-mean. What about non-zero mean signals and/or noise? Can use same derived filter but subtract mean before wiener f
Image Restoration age esto at o
Basic idea: assume the source of the distortion is known; use this information and possibly info about the image to reconstruct the p y g image.
x(n1,n2) degradation
Example: digitizer distortion E l di iti di t ti recorder
Qua t at o Quantization
Some sort of quantization is necessary to represent continuous
signals in digital form
x(n1,n2) x(t1,t2) t 2D Sampler Quantizer xq(n1,n2) n
Digitizer (A/D)
Quantization is also used for data reduction in virtually all lossy
codin
Image Enhancement and Image Restoration age a ce e t a d age esto at o
Image Enhancement
Objective: accentuate or improve appearance of features, for subsequent analysis or display (possibly, but not necessarily degraded by some phenomenon). Examples of
Homomorphic Processing o o o p c ocess g
Motivation: Image with a large dynamic range, e.g. natural
scene on a bright sunny day, recorded on a medium with a small dynamic range, e.g. a film image contrast significantly reduced especially in the dark and
Nonlinear Filtering for Noise S oot g o ea te g o o se Smoothing
Frequency Domain Method for Noise Smoothing
Example: Picture with lines on it. Can we get rid of lines? Think of the image as desired image + noise (lines).
Consider working in transformed
Vision & Perception so e cept o
Simple model:
simple reflectance/illumination model
illumination source i(n1,n2) Eye
image: where
x(n1,n2)=i(n1,n2)r(n1,n2) 0 < i(n1,n2) < 0 < r(n1,n2) < 1
reflectance term r(n1,n2)
EEE 508
Vision & Perception so e cept o
Image Segmentation age Seg e tat o
Objective: to determine (extract) object boundaries. It is a process of partitioning an image into distinct regions by
grouping together neighboring pixels based on some predefined similarity criterion. It can be viewed
EEE 508 - Digital Image Processing and Compression http:/lina.faculty.asu.edu/eee508/ http:/lina faculty asu edu/eee508/
Introduction Prof. Lina Karam P f Li K School of Electrical, Computer, & Energy Engineering Arizona State University karam@asu.edu
Pow
EEE 508 - Digital Image Processing and Compression
2D DSP Basics: 2D Systems
EEE 508 - Lecture 2
2D Systems
x(n1,n2) T[] y(n1,n2) = T [x(n1,n2)]
Linearity
Additivity:
If x2(n1,n2) ( Then x1 (n1,n2) + x2 (n1,n2)
T T
y2(n1,n2) = T [x2 (n1,n2)] y (n1,n2) = y
EEE 508 - Digital Image Processing and Compression
2D DSP Basics: Systems Stability, 2D Sampling
EEE 508 - Lecture 3
Stab ty Stability
System is stable if a bounded input always results in a
bounded output (BIBO)
For LSI system, a sufficient condition fo
Visual sensation
Complex process, can be divided into 4 main stages:
Image formation
Incoming light is transformed by the optics of the eye and focused on the retina to create the retinal image. This is a series of simple and well known optical transfor
Organization in the Visual Cortex
Hypercolumn based structure of the striate cortex
Information in the visual cortex is represented within an orientation- and frequency-selective modular neural bank. A hypercolumn consists of two ocular dominance column
Lateral Inhibition
Copyright 2010 by Lina J. Karam
Lateral Inhibition
Mach Band Effect
Copyright 2010 by Lina J. Karam
Lateral Inhibition
Mach Band Effect Luminance versus brightness
Luminance Gr level ray Brightness
Distance
Copyright 2010 by Lina J. Kar
Super Resolution Super-Resolution
Super-Resolution ( ) i l i (SR) image re-construction is the i i h process of combining the information from multiple Low Resolution Low-Resolution (LR) aliased and noisy frames of the same scene to estimate a High-Resolu