video technology 2013-14 - minor ii
TRANSCRIPT
-
8/12/2019 Video Technology 2013-14 - Minor II
1/282
Image &Video Processing
1
B.Tech Sem VIII
-
8/12/2019 Video Technology 2013-14 - Minor II
2/282
Course Contents
1. Introduction to Image, Video1. Human Visual System (HVS)
2. Colours
Biology, Physics, Technology, Coding
3. Image Processing (examples)
Capture, Preprocessing, 1D and 2D Fourier transformation,
Minor I
1D an D convo ut on, reconstruct on, a as ng, ter ng
Enhancement 1. noise reduction, filter masks
2. edge-detection, histograms,
3. image segmentation4. Image Compression (example)
5. History and Basics of Video Technology (examples)
6. Video Compression
Minor II
-
8/12/2019 Video Technology 2013-14 - Minor II
3/282
MINOR I
-
8/12/2019 Video Technology 2013-14 - Minor II
4/282
Images
Images, often calledpictures, are represented bybitmaps.
A bitmap is a spatial two-dimensional matrix made upof individual picture elements calledpixels.
Each pixel has a numerical value calledamplitude.
4
The number of bits available to code a pixel is calledamplitude depth orpixel depth.
A pixel depth may represent
a black or white dot in images
a level of gray in continuous-tone, monochromatic images, or
the color attributes of the picture element in colored pictures.
-
8/12/2019 Video Technology 2013-14 - Minor II
5/282
An Image is...
(will discuss in later lectures)52 6B 8C 6B 73 5A 635A 6B 73 84 84 73 73
5A 84 84 73 5A 84 84
6B 6B 8C 5A 42 4A 42
42 6B 6B 5A A5 DE D6
5A 6B 5A 42 F7 F7 F7
84 5A 6B 31 DE F7 F7
45 71 82 7D 7D 55 5D
55 75 7D 75 71 6D 65
55 75 7D 6D 61 75 75
made up of pixels
each pixel is a
combination of Red
Green and Blue color
amplitudes
45 71 75 55 41 41 38
34 51 55 55 9E CF CF
51 55 51 38 FF FF FF
71 51 59 34 DB FF FF
42 7B 7B 9C 7B 63 4A
63 7B 7B 73 73 63 63
63 73 7B 63 63 73 7342 73 73 63 42 42 39
31 4A 4A 63 94 DE DE
4A 4A 42 39 FF FF FF
73 42 5A 31 BD FF FF
Represented by I(x,y) of the two spatial coordinates ofthe image plane.
I(x,y) is the intensity of the image at the point (x,y)
on the image plane.
Color image represented by R(x,y), G(x,y), B(x,y)
-
8/12/2019 Video Technology 2013-14 - Minor II
6/282
Image processing An image processing operation typically defines a new
image g in terms of an existing imagef.
The simplest operations are those that transform each
pixel in isolation. These pixel-to-pixel operations can be
written:
g(x,y)t(f(x,y))
Examples: threshold, RGB grayscale
Note: a typical choice for mapping to grayscale is to apply
the YIQ television matrix and keep the Y.
Y
I
Q
0.299 0.587 0.114
0.596 0.275 0.321
0.212 0.523 0.311
R
G
B
-
8/12/2019 Video Technology 2013-14 - Minor II
7/282
Pixel movement
Some operations preserve intensities, but movepixels around in the image
( , ) ( ( , ), ( , ))g x y f x x y y x y
7
Examples: many amusing warps of images
[Show image sequence.]
-
8/12/2019 Video Technology 2013-14 - Minor II
8/282
Recommended Books
Gonzales and R.E Woods, Digital Image Processing, AddisonWesley, 1993, 3rd Edition.
A.K Jain, Digital Image Processing, Prentice Hall India. 1989
K.R. Rao & J.J. Hwang, Techniques & Standards for Image, , , .
-
8/12/2019 Video Technology 2013-14 - Minor II
9/282
Overview: Image Formats Uncompressed
pgm (portable gray map) or ppm (portable pixel map) Unix,
bmp (gary and color) Windows.
Compressed GIF (Graphics Interchange Format) :
Average Compression Ratio 4:1.
GIF87a and GIF 89a.
9
.
JPEG (Joint Photographic Experts Group) Good for photos, not very good for small image or line arts less
than 100x100 pixels. Compression ratio 10:1 to 100:1.
PNG (Portable Network Graphics) More color depth (up to 48bit) than GIF(8bit). 10 30 % smaller than GIF. Automatic anti-alias. Text based metadata can be added.
tif, tiff, ps, pdf, eps etc.
-
8/12/2019 Video Technology 2013-14 - Minor II
10/282
Clarifications: Dimension in different
context Dimension of a signal ~ # of index variables
Audio and speech is 1-D signal: over time or sampled time index
Image is 2-D: over two spatial indices (horizontal and vertical)
Video is 3-D: over two spatial indices and one time index
Dimension of an image ~ size of digital image How man ixels alon each row and column: e. . 512x512 ima e Also referred to as the resolution of an image
Dimension of a vector space ~ # of basis vectors in it [ x(1), , x(N) ]T ~ # of elements in the vector
-
8/12/2019 Video Technology 2013-14 - Minor II
11/282
Colours
Vipan Kakkar 11
-
8/12/2019 Video Technology 2013-14 - Minor II
12/282
Upon completion of this unit, you
should be able to explain the
colours and describe how it is
Colours
Vipan Kakkar 12
.
-
8/12/2019 Video Technology 2013-14 - Minor II
13/282
What is colour?
colour
:
The appearance of objects or light
sources described in terms of the
individuals perception of them,
13
involvinghue,brightness, and
saturation.
colour is a sensation -- a perception
of the viewer.Websters New College Dictionary
-
8/12/2019 Video Technology 2013-14 - Minor II
14/282
The Visible Spectrum
-
8/12/2019 Video Technology 2013-14 - Minor II
15/282
The colour of Light
15
-
8/12/2019 Video Technology 2013-14 - Minor II
16/282
Light
Illuminating sources:
emit light (e.g. the sun, light bulb, TV monitors)
perceived colour depends on the emitted freq.
follows additive rule
R+G+B=White
16
Reflecting sources:
reflect an incoming light (e.g. the colour dye, cloth etc
perceived colour depends on reflected freq (=emitted freq-
absorbed freq.) follows subtractive rule
R+G+B=Black
-
8/12/2019 Video Technology 2013-14 - Minor II
17/282
Wavelength of the light
(RGB)
17
-
8/12/2019 Video Technology 2013-14 - Minor II
18/282
The colour Sensitivity of Light
visible region of the
Spectrum.
This visible region is a very
narrow segment of this
18
extending from
~ 440nm in the extreme
blue (near ultra violet)
to
~ 690 nm in the red
region--with green in the
middle @ ~ 555 nm.
-
8/12/2019 Video Technology 2013-14 - Minor II
19/282
Primary colours
19
Red Green Blue
-
8/12/2019 Video Technology 2013-14 - Minor II
20/282
Food for thought
Why R, G, B?
These are primary colours.
y are ese cons ere pr mary co ours
-
8/12/2019 Video Technology 2013-14 - Minor II
21/282
Food for thought
Why R, G, B?
These are primary colours.
Why are these considered primary colours.
Due to laws of physics
Due to laws of biology
-
8/12/2019 Video Technology 2013-14 - Minor II
22/282
Human Visual System (HVS)
Eyes, optic nerve, parts of the
brain
Transforms electroma netic
energy
-
8/12/2019 Video Technology 2013-14 - Minor II
23/282
Human Visual System
Image Formation
Cornea (focus, curvature, RI), sclera
(nerves till optic nerve), pupil (point for
light),iris (controls the quantity oflight), ens (focus), re na (image),
fovea (central vision, cones)
Transduction
retina, rods, and cones Processing
optic nerve, brain
-
8/12/2019 Video Technology 2013-14 - Minor II
24/282
Human Visual System
From a HVS perspective
the retina is composed of
three kinds of cone cells that
have a high sensitivity to .
These are 630nm (red), 530nm
(green) and 450nm (blue).
So, All image/video displays arebased on this theory of vision to
reproduce color.
-
8/12/2019 Video Technology 2013-14 - Minor II
25/282
The Human Vision System (HVS)
Naturally, an eye can adapt to ahuge range of intensities from
lowest visible light to highest
bearable glare.
The e e uses two t es of discrete
light receptors.
6-7 million centrally located
cones are highly sensitive to
colour and bright light.
75-100 million rods across the
surface of the retina are
sensitive to light but not colour.
Above : A cardiologist examining a coronary angiogram.
Above right: http://www.wiu.edu/users/mfmrb/PSY343/SensPercVision.htm
-
8/12/2019 Video Technology 2013-14 - Minor II
26/282
Transduction (Retina)
Transform light to neural
impulses
Converts light energy into
electrical impulses
Bipolar cells signal
ganglion cells
Axons in the ganglion cells
form optic nerveRodsBipolar cells
ConesGanglion
Optic nerve
-
8/12/2019 Video Technology 2013-14 - Minor II
27/282
The Human Vision System (HVS) As in the camera the image is
projected upside down on theback surface.
However, the human vision system(HVS) is an extremelysophisticated multi-stage process.
The first steps in the sensoryprocess of vision involve thestimulation of light receptors to
.
Electrical signals containing thevision information from each eyeare transmitted to the brainthrough the optic nerves.
The image information is
processed in several stages,ultimately reaching the visualcortex of the cerebrum.
We see with our brain not oureyes!
-
8/12/2019 Video Technology 2013-14 - Minor II
28/282
Rods vs Cones
Contain photo-pigment
Respond to low energy
Enhance sensitivit
Contain photo-pigment
Respond to high energy
Enhance erce tion
Cones Rods
Concentrated in retina, butoutside of fovea
One type, sensitive to
grayscale changes
Concentrated in fovea, existsparsely in retina
Three types, sensitive to
different wavelengths
-
8/12/2019 Video Technology 2013-14 - Minor II
29/282
Camera and Eye
-
8/12/2019 Video Technology 2013-14 - Minor II
30/282
Food for thought
Why R, G, B?
These are primary colours.
Why are these considered primary colours.
Due to laws of physics
Due to laws of biology
Primary colours therefore are primary only becausecones in our eyes are sensitive to those three colours.
-
8/12/2019 Video Technology 2013-14 - Minor II
31/282
Tri-stimulus Theorem (1/3)
RGB Model: Different intensities ofred, green, and blue are added to generate
various colors.
Luma/ Chroma Representation (to be described later):
The luminance component (Y) contains the gray-scale information
31
. ., .
The chrominance component defines the color (U) and theintensity (V) of the color.
Advantage: The human eye is more susceptible to brightness thancolor.
A compression scheme can use gray-scale information to define detailand allows loss of color information to achieve higher rates of compression(i.e., JPEG).
-
8/12/2019 Video Technology 2013-14 - Minor II
32/282
Primary colors cannot be obtained by mixing the other two primary
colors.
Tri-stimulus Theorem (2/3)
32
are three types of color receptors in a human eye.
-
8/12/2019 Video Technology 2013-14 - Minor II
33/282
Tri-stimulus Theorem (2/3)
3 types of cones (6 to 7 million of them) Red = L cones, Green = M cones, Blue = S cones
Ratio differentiates for each person
E.g., Red (64%), Green (32%), rest S cones
E.g., L(75.8%), M(20%), rest S cones
E.g., L(50.6%), M(44.2%), rest S cones Source of information:
See cone cell in wikipedia
www.colorbasics.com/tristimulus/index.php
Each type most responsive to a narrow band red and green absorb most energy, blue the least
Light stimulates each set of cones differently, and the ratiosproduce sensation of color
-
8/12/2019 Video Technology 2013-14 - Minor II
34/282
Color Specification Systems
(Color Spaces)
Spectral Power Distribution (SPD): A plot of radiant energy of a
color vs wavelength.
The luminance, hue (color), and saturation of a color can be
specified most accurately by its SPD. However, SPD does not
describe the relationship between the physical properties of a
34
color and its visual perception.
International Commission on Illumination (CIE-Commission
Internationale dEclairage) system defines how to map an SPD
to a triple-numeral-component that are mathematical
coordinates in a color space (more details during videoprocessing lectures).
-
8/12/2019 Video Technology 2013-14 - Minor II
35/282
What is color space?
A model used to define a specified color
R, G and B represent lighting colors of red, green and blue
respectively.
Colour Space
Combining red, green and blue
- with different weights can produce any visible color.
- A numerical value is used to indicate the proportion of each color.
-Drawback:
-3 colors are equally important and should be stored with same
amount of data bits.
-Therefore another color representation may be required (described
later)
-
8/12/2019 Video Technology 2013-14 - Minor II
36/282
The CIE (Chromaticity System) was established to define an
"average" human observer.
The average human eye is most sensitive to green/yellow light
and least sensitive to reds or blues (slide 11).
The CIE System (1/3)
36
apparatus in order to define a "standard observer". The results are
shown here, and are called "CIE color space".
-
8/12/2019 Video Technology 2013-14 - Minor II
37/282
The CIE System (2/3)
CIE 1931 XYZ system
One of the color spaces
The first mathematically defined color space
37
Three parameter:
X, Y, Z
or Y (brightness), x, y (chroma)
-
8/12/2019 Video Technology 2013-14 - Minor II
38/282
The CIE System (3/3)
CIE Chromaticity
Diagram
Spectral Locus
38
Parameter x, y
-
8/12/2019 Video Technology 2013-14 - Minor II
39/282
Refresher: Color Theory
Exam les
-
8/12/2019 Video Technology 2013-14 - Minor II
40/282
Red Green
colour
40
Blue
-
8/12/2019 Video Technology 2013-14 - Minor II
41/282
Red
colour
41
Red+ Blue
Magenta
Red+ Blue
Magenta
Blue
Magenta
-
8/12/2019 Video Technology 2013-14 - Minor II
42/282
Red GreenYellow
colour
42
Green
+ Red
Yellow
Green
+ Red
Yellow
-
8/12/2019 Video Technology 2013-14 - Minor II
43/282
Green
colour
43
Green+ Blue
Cyan
Green+ Blue
Cyan
Blue
Cyan
-
8/12/2019 Video Technology 2013-14 - Minor II
44/282
Red Green
colour
44
Red
Green
+ Blue
Red
Green
+ Blue
White
Blue
White
-
8/12/2019 Video Technology 2013-14 - Minor II
45/282
Additive Theory
Black radiates no light White (sun) radiates all light
Video is the process of capturing andradiating light, therefore it uses Additive(Light) Theory not Subtractive (Pigment)Theory.
The primary colours in Additive Theory are:
Red ( R )
Green ( G )
Blue ( B )
The primary colours add together tomake white
Light Theory is also called AdditiveTheory.
Light Theory is used in Television, theaterlighting, computer monitors, and videoproduction.
-
8/12/2019 Video Technology 2013-14 - Minor II
46/282
The Colour Wheel
formed:
-
8/12/2019 Video Technology 2013-14 - Minor II
47/282
colour Perception (colour Theory) Hue
distinguishes named colours,e.g., RGB
dominant wavelength of thelight
Saturation
Perceived intensity of a
Hue Scale
O
rigi
how far colour is from a grayof equal intensity
Brightness (lightness) perceived intensity
Saturation
nal
lightness
Source: Wikipedia
-
8/12/2019 Video Technology 2013-14 - Minor II
48/282
The Colour Wheel Colours on the wheel
can be described usingthree parameters:
1. Hue: degrees from 0 to
360
2. Saturation: brightnessor dullness
3. Value: lightness or
darkness
(As suggested by Henry Albert Munsell inAColour Notation, 1905)
-
8/12/2019 Video Technology 2013-14 - Minor II
49/282
The Colour Wheel: Hue Hue or Spectral Colour is
represented as an angle.
Primary Colours: 0 = Red
120 = Green
240 = Blue
Secondary Colours: 60 = Yellow
180 = Cyan
300 = Magenta
-
8/12/2019 Video Technology 2013-14 - Minor II
50/282
The Colour Wheel: Saturation Saturation or Chroma is the
intensity of a colour.
A highly saturated colour isbright and appears closer tothe ed e of the wheel.
A more unsaturated colouris dull.
A colour with no saturation
is achromatic or in the greyscale.
-
8/12/2019 Video Technology 2013-14 - Minor II
51/282
The Colour Wheel: Value"the quality by which we
distinguish a light colour
from a dark one."- Albert Henry Munsell
A Colour Notation 1905
Value represents the luminescent
contrast value between black
and white
-
8/12/2019 Video Technology 2013-14 - Minor II
52/282
The Colour Wheel 3dThree parameters to describe a colour: Hue
Chroma Value
-
8/12/2019 Video Technology 2013-14 - Minor II
53/282
MANY more scientific models based on different colour theory: (Example: Colour
Tree by American artist Henry Albert Munsell from
A Colour Notation, 1905.)
-
8/12/2019 Video Technology 2013-14 - Minor II
54/282
Colour SchemesSystematic ways of selecting colours
Monochromatic
Complimentary
Analo ous
Warm
Cool
Achromatic Chromatic Grays
-
8/12/2019 Video Technology 2013-14 - Minor II
55/282
Colour Schemes: Monochromatic
Monochromatic:One Hue many values of Tintand Shade
Artist: Marc ChagallTitle: Les Amants Sur Le Toit
-
8/12/2019 Video Technology 2013-14 - Minor II
56/282
Colour Schemes: Complementary (notespelling--NOT complimentary)
Complementary: Colours thatare opposite on the wheel.High Contrast
Artist: Paul Cezanne
Title: La Montage Saint VictoireYear: 1886-88
-
8/12/2019 Video Technology 2013-14 - Minor II
57/282
Colour Schemes: Analogous
Analogous: A selection ofcolours that are adjacent.Minimal contrast
Artist: Vincent van Gogh
Title: The IrisYear: 1889
-
8/12/2019 Video Technology 2013-14 - Minor II
58/282
Colour Schemes: Warm
Warm: First half of the wheelgive warmer colours. Thecolours of fire.
Artist: Jan Vermeer
Title: Girl Asleep at a TableYear: 1657
-
8/12/2019 Video Technology 2013-14 - Minor II
59/282
Colour Schemes: Cool
Cool: Second half of the wheelgives cooler colours
Artist: Pablo Picasso
Title: Femme Allonge LisantYear: 1939
-
8/12/2019 Video Technology 2013-14 - Minor II
60/282
Colour Schemes:
Achromatic, Chromatic Grays
Achromatic: Black and white with all the
grays in-between.
Chromatic Grays:Also called neutral
relief. Dull colours, low contrast.
-
8/12/2019 Video Technology 2013-14 - Minor II
61/282
Additive Color Mixing
The mixing oflight
61
Primary: Red, Green, Blue
The complementary color
White means
-
8/12/2019 Video Technology 2013-14 - Minor II
62/282
Subtractive Color Mixing (1/2..)
The mixing ofpigment
62
Primary: Cyan, Magenta, Yellow
The complementary color
Why black?
-
8/12/2019 Video Technology 2013-14 - Minor II
63/282
Subtractive Color Mixing (2/2..)
Why?
Pigments absorb light
63
Thinking:
the Color Filters
Question:
Yellow + Cyan=?
-
8/12/2019 Video Technology 2013-14 - Minor II
64/282
Image Acquisition
-
8/12/2019 Video Technology 2013-14 - Minor II
65/282
Sensor:Charge-coupled device (CCD)
Special sensor that captures an image
Light-sensitive silicon solid-state device composed of many cells
When exposed to light, each
cell becomes electrically
charged. This charge can
then be converted to a 8-bit
The electromechanical
shutter is activated to expose
the cells to light for a brief
Lens
area
65
value where 0 represents no
exposure while 255
represents very intense
exposure of that cell to light.
Some of the columns are
covered with a black strip of
paint. The light-intensity ofthese pixels is used for zero-
bias adjustments of all the
cells.
moment.
The electronic circuitry, when
commanded, discharges the
cells, activates the
electromechanical shutter,
and then reads the 8-bit
charge value of each cell.These values can be clocked
out of the CCD by external
logic through a standard
parallel bus interface.
Pixel
columns
overe
columns
Electronic
circuitry
Electro-
mechanical
shutter
Pixel
rows
-
8/12/2019 Video Technology 2013-14 - Minor II
66/282
Sensor:Charge-coupled device (CCD)
Special sensor that captures an image
Light-sensitive silicon solid-state device composed of many cells
When exposed to light, each
cell becomes electrically
charged. This charge can
then be converted to a 8-bit
The electromechanical
shutter is activated to expose
the cells to light for a brief
Lens
area
66
value where 0 represents no
exposure while 255
represents very intense
exposure of that cell to light.
Some of the columns are
covered with a black strip of
paint. The light-intensity ofthese pixels is used for zero-
bias adjustments of all the
cells.
moment.
The electronic circuitry, when
commanded, discharges the
cells, activates the
electromechanical shutter,
and then reads the 8-bit
charge value of each cell.These values can be clocked
out of the CCD by external
logic through a standard
parallel bus interface.
Pixel
columns
overe
columns
Electronic
circuitry
Electro-
mechanical
shutter
Pixel
rows
-
8/12/2019 Video Technology 2013-14 - Minor II
67/282
Image Sampling And Quantisation
Remember that a digital image is always only
an approximation of a real world scene
-
8/12/2019 Video Technology 2013-14 - Minor II
68/282
An Image is... (1/2)52 6B 8C 6B 73 5A 63
5A 6B 73 84 84 73 73
5A 84 84 73 5A 84 84
6B 6B 8C 5A 42 4A 42
42 6B 6B 5A A5 DE D6
5A 6B 5A 42 F7 F7 F7
84 5A 6B 31 DE F7 F7
45 71 82 7D 7D 55 5D
55 75 7D 75 71 6D 6555 75 7D 6D 61 75 75
made up of pixels
each pixel is a
combination of Red
Green and Blue color
amplitudes
45 71 75 55 41 41 38
34 51 55 55 9E CF CF
51 55 51 38 FF FF FF
71 51 59 34 DB FF FF
42 7B 7B 9C 7B 63 4A
63 7B 7B 73 73 63 63
63 73 7B 63 63 73 73
42 73 73 63 42 42 39
31 4A 4A 63 94 DE DE
4A 4A 42 39 FF FF FF
73 42 5A 31 BD FF FF
Represented by I(x,y) of the two spatial coordinates ofthe image plane.
I(x,y) is the intensity of the image at the point (x,y)
on the image plane. f(x,y) can also be interchangeably used as
A function.
Color image represented by R(x,y), G(x,y), B(x,y)
-
8/12/2019 Video Technology 2013-14 - Minor II
69/282
An Image is... (2/2)
A color image is just three functions pasted together. Wecan write this as a vector-valued function:
( , )
( , ) ( , )
( , )
r x y
f x y g x y
b x y
Well focus in grayscale (scalar-valued) images for now.
-
8/12/2019 Video Technology 2013-14 - Minor II
70/282
Spatial and Frequency Domains
Spatial domain
refers to planar region of
intensity values at time t
Spatial domain Frequency domain
requency oma n
An image plane as a
sinusoidal function of
changing intensity values
refers to organizing pixels
according to their changing
intensity (frequency)
CS 414 - Spring 2009
f(x,y)
F(sx,sy )
-
8/12/2019 Video Technology 2013-14 - Minor II
71/282
-
8/12/2019 Video Technology 2013-14 - Minor II
72/282
Image Transformation
Fourier Transform
-
8/12/2019 Video Technology 2013-14 - Minor II
73/282
Filtering in the Frequency Domain
-
8/12/2019 Video Technology 2013-14 - Minor II
74/282
Fourier transforms
We can represent a function as a linear combination
(weighted sum) of sines and cosines.
We can think of a function in two complementary ways:
Spatially in the spatial domain
The Fourier transform and its inverse convert between these
two domains:
Frequencydomain
Spatialdomain
F(s) f(x)e i2sx
dx
f(x) F(s)e i2sx
ds
-
8/12/2019 Video Technology 2013-14 - Minor II
75/282
2D Fourier transform
Frequency
domain
Spatial
domain
Spatial domain Frequency domain
F(sx,sy ) f(x,y)ei2sxx
ei2syydxdy
f(x,y) F(sx,sy )ei2sxx
e i2syydsxdsy
f(x,y)
F(sx ,sy )
-
8/12/2019 Video Technology 2013-14 - Minor II
76/282
Fourier transforms (contd)
Where do the sines and cosines come in?
Frequency
domain
Spatial
domain
F(s) f(x)e i2
sx
dx
f(x) F(s)ei2sx
ds
f(x) is usually a real signal, but F(s) is generallycomplex:
Iff(x) is symmetric, i.e.,f(x) =f(-x)), then F(s) =Re(s).
F(s) Re(s) i Im(s) F(s) ei2s
-
8/12/2019 Video Technology 2013-14 - Minor II
77/282
What if f(x,y) were separable? That is,
f x = f x f
)(2e),(f),F( dxdyyxSySx ySxSi yx
Fourier transforms (contd)
)(221 e)()(),F( dxdyyfxfSySx ySxSi yx
)(22)(2
1 e)(e)(),F( dxdyyfxfSySx ySixSi xx
Breaking up the exponential,
Fourier transforms (contd)
-
8/12/2019 Video Technology 2013-14 - Minor II
78/282
dyyfdxxfSySx ySixSi yx )(22)(2
1 e)(e)(),F(
Separating the integrals,
Fourier transforms (cont d)
, 21 yxyx Using these two,
-the spatial domain image is first transformed into an intermediate image
using N one-dimensional Fourier Transforms.-This intermediate image is then transformed into the final image, again
using N one-dimensional Fourier Transforms.
-
8/12/2019 Video Technology 2013-14 - Minor II
79/282
-
8/12/2019 Video Technology 2013-14 - Minor II
80/282
2D Fourier transform
Frequency
domain
Spatial
domain
F(sx,sy ) f(x,y)e i2sxx
ei2syydxdy
f(x,y) F(sx,sy )ei2sxx
e i2syydsxdsy
Spatial domain Frequency domain
-The Fourier Transform: used to decompose an
ima e into its sine and cosine com onents.
f(x,y)
F(sx,sy )
-The output of the transformation represents
the image in the frequency domain, while the
input image is the spatial equivalent.
- In the Fourier domain image, each pointrepresents a frequency contained in the spatial
domain image.
-
8/12/2019 Video Technology 2013-14 - Minor II
81/282
Example - revisitedSpatial domain Frequency domain
f(x,y)
F(sx,sy )
The Fourier image has two basic
component ?
-
8/12/2019 Video Technology 2013-14 - Minor II
82/282
Example - revisitedSpatial domain Frequency domain
f(x,y)
F(sx,sy )
The Fourier image is shown in such a
way that the DC-value F(0,0) is
displayed in the center of the image.
The further away from the center an
image point is, the higher is its
corresponding frequency.
-
8/12/2019 Video Technology 2013-14 - Minor II
83/282
Example - revisitedSpatial domain Frequency domain
f(x,y)
F(sx,sy )
The Fourier image is shown in such a
way that the DC-value F(0,0) is
displayed in the center of the image.
The further away from the center an
image point is, the higher is its
corresponding frequency.
We can see that the DC-value is by far the
largest component of the image.
However, the dynamic range of the Fouriercoefficients (i.e. the intensity values in the
Fourier image) is too large as obtained in the
image above.
-
8/12/2019 Video Technology 2013-14 - Minor II
84/282
Why FT
Frequency
domain
Spatial
domain
F(sx,sy ) f(x,y)e i2sxx
ei2syydxdy
f(x,y) F(sx,sy )ei2sxx
e i2syydsxdsy
Spatial domain Frequency domain
f(x,y)
F(sx,sy )
Fourier Transform contain a set of samples
which is large enough to fully describe the
spatial domain image.
The number of frequencies corresponds to the
number of pixels in the spatial domain
image, i.e. the image in the spatial and Fourier
domain are of the same size.
-
8/12/2019 Video Technology 2013-14 - Minor II
85/282
2D - FT
Frequency
domain
Spatial
domain
F(sx,sy ) f(x,y)e i2sxx
ei2syydxdy
f(x,y) F(sx,sy )ei2sxx
e i2syydsxdsy
Spatial domain Frequency domain
Where:
f(x,y)
F(sx,sy )
-f(x,y) is the image in the spatial domain and
- the exponential term is the basis function
corresponding to each point F(sx, sy) in the
Fourier space.
-The equation can be interpreted as: the value
of each point F(sx, sy) is obtained bymultiplying the spatial image with the
corresponding base function and summing the
result.
-
8/12/2019 Video Technology 2013-14 - Minor II
86/282
1D Fourier examples
-
8/12/2019 Video Technology 2013-14 - Minor II
87/282
2D Fourier examples
Spatial Frequency
domain
f(x,y)
F(sx ,sy )
-
8/12/2019 Video Technology 2013-14 - Minor II
88/282
DFT
A
2D Fourier examples
DFT
DFT
0.25 * A
+ 0.75 * B
-
8/12/2019 Video Technology 2013-14 - Minor II
89/282
Summary
We have looked at:
Human visual system
Light and the electromagnetic spectrum
Colors in imaging Image capture
Image sensing and acquisition
Image representation Fourier and spatial domains
Sampling, quantisation and resolution
Next topic: image enhancement techniques
-
8/12/2019 Video Technology 2013-14 - Minor II
90/282
Image Pre-processing
M i i fil i d i i
-
8/12/2019 Video Technology 2013-14 - Minor II
91/282
Motivation: filtering and resizing
Pre-processing
What if we now want to:
smooth an image?
sharpen an image?
shrink an image?
Before we try these operations, lets revisit
and think about images in a moremathematical way
-
8/12/2019 Video Technology 2013-14 - Minor II
92/282
Image Resolution
How many pixels
Spatial resolution
How many shades of grey/colours
92
How many frames per second
Temporal resolution
Nyquists theorem
-
8/12/2019 Video Technology 2013-14 - Minor II
93/282
Nyquists Theorem
A periodic signal can be reconstructed if the
sampling interval is half the period
An object can be detected if two samples span
93
-
8/12/2019 Video Technology 2013-14 - Minor II
94/282
Spatial Resolution
94
n, n/2, n/4, n/8, n/16 and n/32 pixels on a side.
-
8/12/2019 Video Technology 2013-14 - Minor II
95/282
Amplitude Resolution
Humans can see:
About 40 shades of brightness
About 7.5 million shades of colour
95
Depends on signal to noise ratio
40 dB equates to about 20 shades
Images captured:
256 shades
-
8/12/2019 Video Technology 2013-14 - Minor II
96/282
Shades of Grey
96
256, 16, 4 and 2 shades.
T l R l ti ( ill b d
-
8/12/2019 Video Technology 2013-14 - Minor II
97/282
Temporal Resolution (will be done
later)
Nyquists theorem for temporal data
How much does an object move between frames?
Can motion be understood unambiguously?
97
-
8/12/2019 Video Technology 2013-14 - Minor II
98/282
-
8/12/2019 Video Technology 2013-14 - Minor II
99/282
Filtering in the Frequency Domain
-
8/12/2019 Video Technology 2013-14 - Minor II
100/282
Filtering in the Frequency Domain
-
8/12/2019 Video Technology 2013-14 - Minor II
101/282
Some Basic Filters and Their Functions
Multiply all values ofF(u,v) by the filter function (notch filter):
All this filter would do is set F(0,0) to zero (force the average value of
an image to zero) and leave all frequency components of the Fourier
otherwise.1
)2/,2/(),(if0),(
NMvuvuH
trans orm untouc e .
2 l d h
-
8/12/2019 Video Technology 2013-14 - Minor II
102/282
Basic 2D Filters and Their Functions
Lowpass filter
Highpass filter
-
8/12/2019 Video Technology 2013-14 - Minor II
103/282
Convolution
-
8/12/2019 Video Technology 2013-14 - Minor II
104/282
2D Convolution
Lets Try it with Two-Dimensions!
-
8/12/2019 Video Technology 2013-14 - Minor II
105/282
Let s Try it with Two Dimensions!This image exclusively has 32 cycles
in the vertical direction.
This image exclusively has 8 cycles in
the horizontal direction.
So what is going on here?
The u axis runs from left to right and it represents
the horizontal component of the frequency. The v
axis runs up and down and it corresponds tovertical components of the frequency.
x-y coordinate system
Fourier Transform
You will notice that the second example is a
little more smeared out. This is because the
lines are more blurred so more sine waves are
required to build it. The transform is weighted
so brighter spots indicate sine waves more
frequently used.
The central dot is an average of all the sine waves
so it is usually the brightest dot and used as a
point of reference for the rest of the points.
Since this is inverse space, dots close to the origin
will be further apart in real space than dots that
are far apart on the Fourier Transform. (Again
keeping in mind that these dots refer to the
frequency of a component wave.)
u-v coordinate system
Magnitude vs. Phase
-
8/12/2019 Video Technology 2013-14 - Minor II
106/282
Magnitude vs. Phase
The Fourier Transform is defined as:
Since Computers dont like infinite integrals a Fast Fourier
Transform makes it simpler:
Where F(w) is original function and f(t) is the transformed function
N
yvxui
eyxFvuf
)**(2*
),(),(
These two images are shifted pi with respect tox y
Where F(x,y) is real and f(u,v) is complex.
So what do we do with this?Well instead of representing the complex numbers as real and
imaginary parts we can represent it as Magnitude and Phase
where they are defined as:
Re
Imarctan)(
ImRe)( 22
fPhase
fMagnitude
Magnitude is telling how much of a certain frequency
component is in the image.
Phase is telling where that certain frequency lies in the
image.
each other.
-
8/12/2019 Video Technology 2013-14 - Minor II
107/282
Examples
Fourier
Convolution
-
8/12/2019 Video Technology 2013-14 - Minor II
108/282
Fourier Transform
-
8/12/2019 Video Technology 2013-14 - Minor II
109/282
2D-Fourier Transform
-
8/12/2019 Video Technology 2013-14 - Minor II
110/282
Fourier Transform
-
8/12/2019 Video Technology 2013-14 - Minor II
111/282
Fourier Transform
-
8/12/2019 Video Technology 2013-14 - Minor II
112/282
Image Pre-processing (2D Convolution)
-
8/12/2019 Video Technology 2013-14 - Minor II
113/282
Convolution
One of the most common methods for filtering afunction is called convolution.
In 1D, convolution is defined as:
x x * h x
where
h(x)h(x)
f( x)h(x x)d x
f( x)
h( x x)d x
-
8/12/2019 Video Technology 2013-14 - Minor II
114/282
Convolution properties
Convolution exhibits a number of basic, butimportant properties.
Commutativity:
a(x) b(x)b(x) a(x)
Associativity:
Linearity:
[a(x) b(x)] c(x)a(x) [b(x) c(x)]
a(x) [k b(x)] k [a(x) b(x)]
a(x) (b(x) c(x)) a(x) b(x) a(x) c(x)
-
8/12/2019 Video Technology 2013-14 - Minor II
115/282
Convolution in 2D
In two dimensions, convolution becomes:
g(x,y) f(x,y) h(x,y)
f( x, y)h(x x)(y y)d x d y
where
* =
f(x,y) h(x,y) g(x,y)
h(x,y)h(x,y)
,
l
-
8/12/2019 Video Technology 2013-14 - Minor II
116/282
Discrete convolution
For a digital signal, we define discrete convolution as:
g[i] f[i] h[i]
f[i]h[i i]
116
where
i
f[i]
h[i i]i
h[i] h[i]
l h l
-
8/12/2019 Video Technology 2013-14 - Minor II
117/282
1D convolution theorem example
i l i i 2
-
8/12/2019 Video Technology 2013-14 - Minor II
118/282
Discrete convolution in 2D
Similarly, discrete convolution in 2D becomes:
g[i,j] f[i,j] h[i,j]
f[i , j]h[i i ,j j]
i
118
where
f[i , j]
h[i i, j j]j
i
h[i,j] h[i,j]
2D l i h l
-
8/12/2019 Video Technology 2013-14 - Minor II
119/282
2D convolution theorem example
*
f(x,y) |F(sx,sy)|
h(x,y)
g(x,y)
|H(sx,sy)|
|G(sx,sy)|
R i fil i 2D
-
8/12/2019 Video Technology 2013-14 - Minor II
120/282
Reconstruction filters in 2D
We can perform reconstruction in 2D
Example problem
Find the Fourier transform of
-
8/12/2019 Video Technology 2013-14 - Minor II
121/282
Find the Fourier transform of
Example problem: Answer.
Find the Fourier transform of
-
8/12/2019 Video Technology 2013-14 - Minor II
122/282
Find the Fourier transform of
f(x) = (x/4) (x/2) + .5(x)
Using the Fourier transforms of and and the linearity and scaling properties,
F(u) = 4sinc(4u) - 2sinc2(2u) + .5sinc2(u)
Example problem: Alternative Answer.
Find the Fourier transform of
-
8/12/2019 Video Technology 2013-14 - Minor II
123/282
Find the Fourier transform of
f(x) = (x/4) 0.5(((x/3) * (x))
*
Using the Fourier transforms of and and the linearity and scaling and convolution properties ,
F(u) = 4sinc(4u) 1.5sinc(3u)sinc(u)
2 1 0 1 2 1 -.5 0 .5 1
Plane waves
Lets get an intuitive feel for the plane wave )(2e vyuxi
-
8/12/2019 Video Technology 2013-14 - Minor II
124/282
Let s get an intuitive feel for the plane wave e
The period; the distance betweensuccessive maxima of the waves
defines the direction
Lines of constant phase undulation
in the complex plane
of the undulation.
Plane waves: sine and cosine waves
-
8/12/2019 Video Technology 2013-14 - Minor II
125/282
sin(2**x)
cos(2**x)
Plane waves: sine waves in the complex plane.
-
8/12/2019 Video Technology 2013-14 - Minor II
126/282
sin(10**x)
sin(10**x +4*pi*y)
Two-Dimensional Fourier Transform
-
8/12/2019 Video Technology 2013-14 - Minor II
127/282
Where in f(x,y), xand yare real, not complex variables.
)(2e),(f),F( dxdyyxvu vyuxi
Two-Dimensional Fourier Transform:
Two-Dimensional Inverse Fourier Transform:
)(2e),F(),( dudvvuyxf vyuxi
amplitude basis functions
and phase of
required basis functions
Separable Functions
-
8/12/2019 Video Technology 2013-14 - Minor II
128/282
What if f(x,y) were separable? That is,
f x = f x f
)(2e),(f),F( dxdyyxvu vyuxi
Two-Dimensional Fourier Transform:
)(221 e)()(),F( dxdyyfxfvu vyuxi
)(22)(2
1 e)(e)(),F( dxdyyfxfvu vyiuxi
Breaking up the exponential,
Separable Functions
-
8/12/2019 Video Technology 2013-14 - Minor II
129/282
)(22)(2
1 e)(e)(),F( dxdyyfxfvu vyiuxi
dyyfdxxfvu vyiuxi )(22)(2
1 e)(e)(),F(
Separating the integrals,
)()(),( 21 vFuFvuF
-
8/12/2019 Video Technology 2013-14 - Minor II
130/282
Fourier Transform
f(x,y) = cos(10x)*1
F(u,v) = 1/2 [(u+5,0) +(u-5,0)]
u
v
v
-0.5
0
0.5
Real [F(u,v)]
u
vImaginary [F(u,v)]
-
8/12/2019 Video Technology 2013-14 - Minor II
131/282
-
8/12/2019 Video Technology 2013-14 - Minor II
132/282
Fourier Transform
f(x,y) = sin(40x)
F(u,v) = i/2 [(u+20,0) - (u-20,0)]
u
v
v
-0.5
0
0.5
Real [F(u,v)]
u
vImaginary [F(u,v)]
-
8/12/2019 Video Technology 2013-14 - Minor II
133/282
Fourier Transform
f(x,y) = sin(20x + 10y)
F(u,v) = i/2 [(u+10,v+5) - (u-10,v-5)]
u
v
v
-0.5
0
0.5
Real [F(u,v)]
u
vImaginary [F(u,v)]
Reconstruction filters in 2D
-
8/12/2019 Video Technology 2013-14 - Minor II
134/282
Reconstruction filters in 2D
We can perform reconstruction in 2D
-
8/12/2019 Video Technology 2013-14 - Minor II
135/282
Image Pre-processing (Sampling)
Image Sampling And Quantisation (cont )
-
8/12/2019 Video Technology 2013-14 - Minor II
136/282
Image Sampling And Quantisation (cont)
Remember that a digital image is always only
an approximation of a real world scene
Sampling
-
8/12/2019 Video Technology 2013-14 - Minor II
137/282
Now, we can talk about sampling.
Sampling
The Fourier spectrum gets replicatedby spatial sampling!
How do we recover the signal?
Reconstruction filters
-
8/12/2019 Video Technology 2013-14 - Minor II
138/282
Reconstruction filters
The sinc filter, while ideal, has two drawbacks:
It has large support (slow to compute)
It introduces ringing in practice
We can choose from many other filters
Reconstruction filters
-
8/12/2019 Video Technology 2013-14 - Minor II
139/282
Reconstruction filters
The sinc filter, while ideal, has two drawbacks:
It has large support (slow to compute)
It introduces ringing in practice
We can choose from many other filters
Reconstruction filters in 2D
-
8/12/2019 Video Technology 2013-14 - Minor II
140/282
Reconstruction filters in 2D
We can also perform reconstruction in 2D
-
8/12/2019 Video Technology 2013-14 - Minor II
141/282
MINOR I
-
8/12/2019 Video Technology 2013-14 - Minor II
142/282
Image Pre-processing (Aliasing)
Aliasing
-
8/12/2019 Video Technology 2013-14 - Minor II
143/282
Aliasing
Sampling rate is too low
Aliasing
-
8/12/2019 Video Technology 2013-14 - Minor II
144/282
Aliasing
What if we go below the Nyquist frequency?
Anti-aliasing
-
8/12/2019 Video Technology 2013-14 - Minor II
145/282
Anti aliasing
Anti-aliasing is the process ofremoving the frequencies
before they alias.
Anti-aliasing by analytic prefiltering
-
8/12/2019 Video Technology 2013-14 - Minor II
146/282
We can fill the magic box with analytic pre-filteringof the signal:
Why may this not generally be possible?
-
8/12/2019 Video Technology 2013-14 - Minor II
147/282
MINOR II
-
8/12/2019 Video Technology 2013-14 - Minor II
148/282
Image processing/Enhancement (Noise reduction)
What Is Image Enhancement?
-
8/12/2019 Video Technology 2013-14 - Minor II
149/282
What Is Image Enhancement?
Image enhancement is the process of making
images more useful
The reasons for doing this include:
g g ng n eres ng e a n mages
Removing noise from images
Making images more visually appealing
NoiseI i l i i i f d i bl b bl f
-
8/12/2019 Video Technology 2013-14 - Minor II
150/282
In signal processing, it is often desirable to be able to perform somekind of noise reduction on an image or signal.
Image processing is also useful for noise reduction and edgeenhancement. We will focus on these applications for the remainder ofthe lecture
Common types of noise:
Salt and pepper noise:contains random
150
occurrences of black andwhite pixels
Impulse noise: containsrandom occurrences ofwhite pixels
Gaussian noise:
variations in intensity drawnfrom a Gaussian normaldistribution
Ideal noise reduction
-
8/12/2019 Video Technology 2013-14 - Minor II
151/282
Ideal noise reduction
151
Ideal noise reduction
-
8/12/2019 Video Technology 2013-14 - Minor II
152/282
Ideal noise reduction
152
Practical noise reduction
-
8/12/2019 Video Technology 2013-14 - Minor II
153/282
Practical noise reduction
How can we smooth away noise in a single image?
153
Example revisited: 2D Convolution
k ( filt )
-
8/12/2019 Video Technology 2013-14 - Minor II
154/282
mask (average filter)
Effect of average filters
-
8/12/2019 Video Technology 2013-14 - Minor II
155/282
g
155
Median Filtering
-
8/12/2019 Video Technology 2013-14 - Minor II
156/282
Median Filtering
The median filter is another digital filtering technique, often used to remove
noise. Median filtering is very widely used in digital image processing because it
preserves edges while removing noise.
Median filters
-
8/12/2019 Video Technology 2013-14 - Minor II
157/282
It replaces the value of the center pixel with the median of the intensityvalues in the neighborhood of that pixel.
Median filtering is an operation often used in image processing to reduce
"salt and pepper" noise. A median filter is more effective than convolution
when the goal is to simultaneously reduce noise and preserve edges.
Median filters are particularly effective in the presence ofimpulse noise,
also called salt and pepper noise because of its appearance as white and
black dots superimposed on an image.
For every pixel, a 3x3 neighborhood with the pixel as center is considered.
In median filtering, the value of the pixel is replaced by the median of the
pixel values in the 3x3 neighborhood.
Median Filtering Median filtering is useful for removing
-
8/12/2019 Video Technology 2013-14 - Minor II
158/282
Median filtering is useful for removingnoise but usefully preserves edges.
The median is the central value in arange
Median {4,2,0,1,3,0,5} = ?
Median filtering is a popular low-passfiltering method. Pixel values are sortedand the median (middle value) is output.
Median filtering removes sparse outliers.
Sparse outliers appear as salt andpepper noise in images, i.e., dark pixelsin light areas and light pixels in darkareas. This type of noise was commonin old televisions.
You may use some simple filters in thelaboratory (using matlab). A medianfilter will be used to remove noise.
Passing a 3x3 median filter over
the image pixels shown above
on the right produces the
output on the right.
Notice how the outlier (the 150)
is removed.
Median Filtering Median filtering is useful for removing
-
8/12/2019 Video Technology 2013-14 - Minor II
159/282
Median filtering is useful for removingnoise but usefully preserves edges.
The median is the central value in arange
Median {4,2,0,1,3,0,5} = 2
Median filtering is a popular low-passfiltering method. Pixel values are sortedand the median (middle value) is output.
Median filtering removes sparse outliers.
Sparse outliers appear as salt andpepper noise in images, i.e., dark pixelsin light areas and light pixels in darkareas. This type of noise was common inanalogue television.
You will use some simple filters in thelaboratory. A median filter will be usedto remove noise.
Passing a 3x3 median filter over
the image pixels shown above
on the right produces the
output on the right.
Notice how the outlier (the 150)
is removed.
Effect of median filters
-
8/12/2019 Video Technology 2013-14 - Minor II
160/282
160
Mean and Median Filtering
-
8/12/2019 Video Technology 2013-14 - Minor II
161/282
g
X1 X2 X3
X4 X0 X5
X6 X7 X8
X1 X2 X3
X4 X0 X5
X6 X7 X8
Replace the X0 by the
mean of X0~X8 is
called mean filtering
Replace the X0 by the
median of X0~X8 is
called median filtering
Gaussian filters
-
8/12/2019 Video Technology 2013-14 - Minor II
162/282
Gaussian filters weigh pixels based on their distance fromthe center of the convolution filter. In particular:
2 2 2( ) /(2 )
[ , ]i j
eh i j
C
162
This does a decent job of blurring noise while preservingfeatures of the image.
What parameter controls the width of the Gaussian?
What happens to the image as the Gaussian filter kernel
gets wider? What is the constant C? What should we set it to?
Effect of Gaussian filters
-
8/12/2019 Video Technology 2013-14 - Minor II
163/282
163
Comparison: Gaussian noise
-
8/12/2019 Video Technology 2013-14 - Minor II
164/282
p
164
Comparison: salt and pepper noise
-
8/12/2019 Video Technology 2013-14 - Minor II
165/282
165
-
8/12/2019 Video Technology 2013-14 - Minor II
166/282
Image processing/Enhancement (Edge Detection)
Edge detection
-
8/12/2019 Video Technology 2013-14 - Minor II
167/282
One of the most important uses of imageprocessing is edge detection:
Really easy for humans
Really difficult for computers
167
Fundamental in computer vision
Important in many graphics applications
What is an edge?
-
8/12/2019 Video Technology 2013-14 - Minor II
168/282
168
Q: How might you detect an edge in 1D?
Gradients
-
8/12/2019 Video Technology 2013-14 - Minor II
169/282
The gradient is the 2D equivalent of the derivative:
( , ) ,f f
f x yx y
169
Properties of the gradient
Its a vector
Points in the direction of maximum increase off
Magnitude is rate of increase How can we approximate the gradient in a discrete image?
Less than ideal edges
-
8/12/2019 Video Technology 2013-14 - Minor II
170/282
170
Edge Properties
-
8/12/2019 Video Technology 2013-14 - Minor II
171/282
Edge has twoproperties:
how steep it is
direction, ie, is it
A(x)
x
or right?
A(x)
x
Edge PropertiesEdge Properties gradientgradient
-
8/12/2019 Video Technology 2013-14 - Minor II
172/282
Consider a 1-dcontinuous image of
an edge, denoted by
Edge PropertiesEdge Properties--gradientgradient
0
100
200
A(x)
Edge properties canbe obtained from the
gradient = A /x
gradient=dA/dxasx0.
5 10 15 20 25 30-20
0
20
5 10 15 20 25 30-20
0
20
xA
A
x
Edge Properties
-
8/12/2019 Video Technology 2013-14 - Minor II
173/282
Gradient has two
properties magnitude direction
dx
dAgradient 11
dx
dAgradient 22
Magnitude, or
steepness, given by
|dA/dx|
Direction, left or
right, given by sign ofdA/dx
dxdA
dxdA
dx
dA
dx
dA
12
12
sgnsgn
Edge PropertiesEdge Properties--gradientgradient
-
8/12/2019 Video Technology 2013-14 - Minor II
174/282
Gradient given by firstderivative dA /d x.
Second derivative,
dge Propertiesdge Properties gradientgradient
5 10 15 20 25 300
100
200 xA
x ,generatestwo peaks at
beginning and end of
edge.
Called ringing.
5 10 15 20 25 30-20
0
20
5 10 15 20 25 30-20
0
20
2
2
dx
Ad
dx
Edge Properties-discrete gradient
-
8/12/2019 Video Technology 2013-14 - Minor II
175/282
5 10 15 20 25 30
0
100
200
20
iAxA B=[-1 1]
5 10 15 20 25 30-20
0
5 10 15 20 25 30-20
0
20
iiii
iiii
ii
iii
AAAA
AAAA
AAdx
Ad
AAAdx
2
12
112
12
2
1
2
B=[1 -2 1]
Steps in edge detection
-
8/12/2019 Video Technology 2013-14 - Minor II
176/282
Edge detection algorithms typically proceedin three or four steps:
Filtering: cut down on noise
Enhancement: amplify the difference between
176
e ges an non-e ges Detection: use a threshold operation
Localization (optional): estimate geometry ofedges beyond pixels
22--d Gradient Operatord Gradient Operator
-
8/12/2019 Video Technology 2013-14 - Minor II
177/282
dy
dA
dx
dAA
22
ji
j
dA/dy
dxdA
dydA
dydx
1-
tannOrientatio
Magn tu e
i
x
Review
-
8/12/2019 Video Technology 2013-14 - Minor II
178/282
Image Enhancement / (Pre)Processing
Noise reduction
Information Detection
Histograms
Image Segmentation
Image Compression (Next)
Review
-
8/12/2019 Video Technology 2013-14 - Minor II
179/282
Image Enhancement / (Pre)Processing
Noise reduction
Using Filters (coefficients of the filter mask (hi,j) to be
computed ???)
Information Detection
Edge Detection
Histograms
Image SegmentationThreshold technique
Image Compression (Next)
Review
-
8/12/2019 Video Technology 2013-14 - Minor II
180/282
Image Enhancement / (Pre)ProcessingNoise reduction
Using Filters (coefficients of the filter mask to be computedusing rectangular, gaussian, triangular techniques etc.)
Avera e Filter mask
Median Filter
Gaussian Filter etc.
Information DetectionEdge Detection
HistogramsImage Segmentation
Threshold technique
-
8/12/2019 Video Technology 2013-14 - Minor II
181/282
-
8/12/2019 Video Technology 2013-14 - Minor II
182/282
Image gradient The gradient of an image:
-
8/12/2019 Video Technology 2013-14 - Minor II
183/282
The gradient points in the direction of most rapid change in intensity
The gradient direction is given by:
how does this relate to the direction of the edge?
The edge strength/magnitude is given by the gradient magnitude
Neighbourhood Operators
-
8/12/2019 Video Technology 2013-14 - Minor II
184/282
First derivative can be calculated by
convolving with mask ______
Second derivative can be calculated by
________
Edge Properties-discrete gradient
-
8/12/2019 Video Technology 2013-14 - Minor II
185/282
5 10 15 20 25 30
0
100
200
20
iAxA B=[-1 1]
take derivative
5 10 15 20 25 30-20
0
5 10 15 20 25 30-20
0
20
iiii
iiii
ii
iii
AAAA
AAAA
AAdx
Ad
AAAdx
2
12
112
12
2
1
2
B=[1 -2 1]
Neighbourhood Operators
-
8/12/2019 Video Technology 2013-14 - Minor II
186/282
First derivative can be calculated by
convolving with mask B=[-1 1].
Second derivative can be calculated by
- .
Discrete 2-d gradient operator
-
8/12/2019 Video Technology 2013-14 - Minor II
187/282
jiijiji
ji
AAAx
A
AyxA
,,,1
,,
1
Neighbour hood
operators
ji jijjiiji
jijjiji
AAA
AAAy
A
,,,
,,1,
11
1
j
i
B
Gradient Operators for Images
-
8/12/2019 Video Technology 2013-14 - Minor II
188/282
Second-order gradient
denoted by 2A.
in an image.
Scalar. 22
2
yxA
Laplacian Operator
Neighbourhood Operators
-
8/12/2019 Video Technology 2013-14 - Minor II
189/282
jijijiji
jiijiijii
ji
AAAA
AAAx
A
AyxA
,,1,1,2
,
2
,,12
2
,,
1
2
ii BAA
jijiji
jijijiji
jijjijjij
jijiji
AAA
AAAA
AAAy
A
AAA
,1,2,
,1,1,2,
,
2
,1,2
2
,,1,2
2
2
121
1
2
j
jj
B
BAA
Neighbourhood Operators
-
8/12/2019 Video Technology 2013-14 - Minor II
190/282
222
ji
ji
BABA
AAA
010
141
010
ji BBB
B*A
What is Edge-enhancement?
-
8/12/2019 Video Technology 2013-14 - Minor II
191/282
Physcophysical experiments indicate that an
image with accentuated or crispened edges is
often more subjectively pleasing than the original
image.
Edge enhancement
-
8/12/2019 Video Technology 2013-14 - Minor II
192/282
Laplacian
Canny edge detector
Laplacian Image
-
8/12/2019 Video Technology 2013-14 - Minor II
193/282
2AL
010
141B
LaplacianLaplacian
-
8/12/2019 Video Technology 2013-14 - Minor II
194/282
Add Laplacian to
2Ad
xA5 10 15 20 25 30
0
100
200
20
. A(x)+L
Overshoot below
and above edge.
22
2
2
2
dx
AdxA
dx
Ad
dx
5 10 15 20 25 30-20
0
5 10 15 20 25 30-20
0
20
0 5 10 15 20 25 30
100
200
300
Neighbourhood Operations
-
8/12/2019 Video Technology 2013-14 - Minor II
195/282
010000
, LyxA
Laplacian :
mask
010151
010
010000
This is enhanced mask BB
(A FILTER)(A FILTER)
Laplacian
-
8/12/2019 Video Technology 2013-14 - Minor II
196/282
A(x)+L
Edge
Enhancement
x
-
8/12/2019 Video Technology 2013-14 - Minor II
197/282
Original imageOriginal image enhanced with
laplacian
Edge detection
-
8/12/2019 Video Technology 2013-14 - Minor II
198/282
original
Edge detector
-
8/12/2019 Video Technology 2013-14 - Minor II
199/282
thinning
(non-maximum suppression)
Effect of (Gaussian kernel size)
-
8/12/2019 Video Technology 2013-14 - Minor II
200/282
Canny with Canny withoriginal
The choice of depends on desired behavior
large detects large scale edges
small detects fine features
Gaussian Mask
-
8/12/2019 Video Technology 2013-14 - Minor II
201/282
Uses Gaussian Mask
Sigma is a value chosen by DESIGNER
22 2/ xe
X and Y are the distances awayfrom target pixel
Range from positive to negative
Mask Radius
Example: Gaussian
-
8/12/2019 Video Technology 2013-14 - Minor II
202/282
-
8/12/2019 Video Technology 2013-14 - Minor II
203/282
Image Enhancement (Histogram)
Image Histograms
-
8/12/2019 Video Technology 2013-14 - Minor II
204/282
The histogram of an image shows us thedistribution of grey levels in the image
Massively useful especially in segmentation
Grey Levels
Freq
uencies
Histogram Examples (cont)
-
8/12/2019 Video Technology 2013-14 - Minor II
205/282
talImageProcessing(2002)
Imagestaken
fromGonzalez&Woods,Dig
i
Histogram Examples (cont)
-
8/12/2019 Video Technology 2013-14 - Minor II
206/282
talImageProcessing(2002)
Imagestaken
fromGonzalez&Woods,Digi
Histogram Examples (cont)
-
8/12/2019 Video Technology 2013-14 - Minor II
207/282
talImageProcessing(2002)
Imagestaken
fromGonzalez&Woods,Digi
Histogram Examples (cont)
-
8/12/2019 Video Technology 2013-14 - Minor II
208/282
talImageProcessing(2002)
Imagestaken
fromGonzalez&Woods,Digi
Histogram Examples (cont)
-
8/12/2019 Video Technology 2013-14 - Minor II
209/282
A selection of images andtheir histograms
Notice the relationships
talImageProcessing(2002)
their histograms
Note that the high contrast
image has the most
evenly spaced histogramImagestaken
fromGonzalez&Woods,Digi
VARIANCE and STANDARD DEVIATION
-
8/12/2019 Video Technology 2013-14 - Minor II
210/282
of the histogram tell us about theaverage contrast of the image !
So, from previous figure:
higher the VARIANCE = higher the
STANDARD DEVIATION,
higher will be the images contrast !
-
8/12/2019 Video Technology 2013-14 - Minor II
211/282
Image Enhancement (Image Segmentation)
Image Segmentation
-
8/12/2019 Video Technology 2013-14 - Minor II
212/282
Segmentation divides an image into itsconstituent regions or objects.
Segmentation of images is a difficult task in
. . Segmentation allows to extract objects in
images.
What it is useful for
-
8/12/2019 Video Technology 2013-14 - Minor II
213/282
segmenting the image, gives the contours of objects, whichcan be extracted using edge detection and/or border
following techniques.
Shape of objects can be described.
, , . Image segmentation techniques are extensively used in
similarity searches, e.g.:
http://elib.cs.berkeley.edu/photos/blobworld/
Segmentation Algorithms
-
8/12/2019 Video Technology 2013-14 - Minor II
214/282
Segmentation algorithms are based on one of twobasic properties of color, gray values, or texture:discontinuity and similarity.
First category is to partition an image based on
,image.
Second category are based on partitioning an imageinto regions that are similar according to apredefined criteria. Histogram thresholding approachfalls under this category.
Clustering in Color Space
-
8/12/2019 Video Technology 2013-14 - Minor II
215/282
1. Each image point is mapped to a point in a color space,e.g.:
Color(i, j) = (R (i, j), G(i, j), B(i, j))
It is many to one mapping.
2. The points in the color space are grouped to clusters.
3. The clusters are then mapped back to regions in the image.
Displaying objects in the Segmented Image
-
8/12/2019 Video Technology 2013-14 - Minor II
216/282
The objects can be distinguished by assigningan arbitrary pixel value or average pixel value
to the pixels belonging to the same clusters.
Thus, one needs clustering algorithms
for image segmentation.
-
8/12/2019 Video Technology 2013-14 - Minor II
217/282
Homework (still preparing):
Implement in Matlab and test on some example images the
clustering in the color space.
Use Euclidean distance in RGB color space.
You can use k-means, PAM, or some other clustering algorithm.Links to k-means, PAM, data normalization
Test images: rose, plane, car, tiger, landscape
-
8/12/2019 Video Technology 2013-14 - Minor II
218/282
Gray Scale Image Example
-
8/12/2019 Video Technology 2013-14 - Minor II
219/282
Image of a Finger Print with light background
Histogram
-
8/12/2019 Video Technology 2013-14 - Minor II
220/282
Segmented Image
-
8/12/2019 Video Technology 2013-14 - Minor II
221/282
Image after Segmentation
Thresholding Bimodal Histograms
-
8/12/2019 Video Technology 2013-14 - Minor II
222/282
Basic Global Thresholding:1)Select an initial estimate for T
2)Segment the image using T. This will produce two groups ofpixels. G1 consisting of all pixels with gray level values >T andG2 consisting of pixels with values
-
8/12/2019 Video Technology 2013-14 - Minor II
223/282
Image of rice with black background
-
8/12/2019 Video Technology 2013-14 - Minor II
224/282
Basic Adaptive Thresholding:
Images having uneven illumination makes it difficult to
-
8/12/2019 Video Technology 2013-14 - Minor II
225/282
g gsegment using histogram,this approach is to divide the original image
into sub images
and use the thresholding process
to each of the sub ima es.
Multimodal Histogram
-
8/12/2019 Video Technology 2013-14 - Minor II
226/282
If there are three or more dominant modes in theimage histogram, the histogram has to be partitionedby multiple thresholds.
u t eve t res o ng c ass es a po nt x,y asbelonging to one object class
if T1 < (x,y) T2and to the background
if f(x,y)
-
8/12/2019 Video Technology 2013-14 - Minor II
227/282
227
Low noise image Thresholded at T=some value
Greylevel thresholding
-
8/12/2019 Video Technology 2013-14 - Minor II
228/282
0.02
p(x)
228
0.00
0.01
x
T
Object
Background
Greylevel thresholding
-
8/12/2019 Video Technology 2013-14 - Minor II
229/282
Comparison with different thresholding
229
High noise circle image Optimum threshold
Relaxation - 20 i
Example: Image Enhancement
-
8/12/2019 Video Technology 2013-14 - Minor II
230/282
Example: Image Enhancement
-
8/12/2019 Video Technology 2013-14 - Minor II
231/282
-
8/12/2019 Video Technology 2013-14 - Minor II
232/282
Assi nment
Color Image
-
8/12/2019 Video Technology 2013-14 - Minor II
233/282
Given Image
Statement
-
8/12/2019 Video Technology 2013-14 - Minor II
234/282
To perform an imagesegmentation:
1. Skin color
Assignment to besubmitted:
latest by March 28,
.
Tools to be used:
Matlab
Report:
Matlab code
Output images
(results) with captions
Example result: Segmented Image
-
8/12/2019 Video Technology 2013-14 - Minor II
235/282
Example result: Segmented image, skin color is shown
-
8/12/2019 Video Technology 2013-14 - Minor II
236/282
Assi nment
Review
h / ( )
-
8/12/2019 Video Technology 2013-14 - Minor II
237/282
Image Enhancement / (Pre)ProcessingNoise reduction
Using Filters (coefficients of the filter mask to be computed usingrectangular, gaussian, triangular techniques etc.)
Average Filter (mask)
Median Filter
.
Information DetectionEdge Detection (to get salient features)
High Pass Filters
Laplacian mask
HistogramsDistribution of intensity levels
Image Segmentation (to extract different objects)Threshold technique
Next: Image Compression
-
8/12/2019 Video Technology 2013-14 - Minor II
238/282
Image Compression
Again: An Image is...
52 6B 8C 6B 73 5A 63
5A 6B 73 84 84 73 73
5A 84 84 73 5A 84 84
made up of pixels
-
8/12/2019 Video Technology 2013-14 - Minor II
239/282
5A 84 84 73 5A 84 84
6B 6B 8C 5A 42 4A 42
42 6B 6B 5A A5 DE D6
5A 6B 5A 42 F7 F7 F7
84 5A 6B 31 DE F7 F7
45 71 82 7D 7D 55 5D
55 75 7D 75 71 6D 65
55 75 7D 6D 61 75 75
each pixel is a
combination of Red
Green and Blue color
amplitudes
45 71 75 55 41 41 38
34 51 55 55 9E CF CF
51 55 51 38 FF FF FF
71 51 59 34 DB FF FF
42 7B 7B 9C 7B 63 4A
63 7B 7B 73 73 63 63
63 73 7B 63 63 73 73
42 73 73 63 42 42 39
31 4A 4A 63 94 DE DE
4A 4A 42 39 FF FF FF
73 42 5A 31 BD FF FF
Represented by I(x,y) of the two spatial coordinates of
the image plane.
I(x,y) is the intensity of the image at the point (x,y)on the image plane.
Color image represented by R(x,y), G(x,y), B(x,y)
Image Compression: JPEG
S S
-
8/12/2019 Video Technology 2013-14 - Minor II
240/282
Summary: JPEG Compression
DCT
Quantization
Zi -Za Scan
Sources: The JPEG website:
http://www.jpeg.org
240
RLE and DPCM
Entropy Coding
Why Compression?
The compression ratio of lossless methods (e.g., Huffman,Arithmetic, LZW) is not high enough for image and video
i
-
8/12/2019 Video Technology 2013-14 - Minor II
241/282
compression. JPEG uses transform coding, it is largely based on the
following observations:
Observation 1: A large majority of useful image contents changerelatively slowly across images, i.e., it is unusual for intensity
241
va ues to a ter up and down severa times in a sma area, orexample, within an 8 x 8 image block.A translation of this fact into the spatial frequency domain,implies, generally, lower spatial frequency components containmore information than the high frequency components whichoften correspond to less useful details and noises.
Observation 2: Experiments suggest that humans are moreimmune to loss of higher spatial frequency components than lossof lower frequency components.
Entropy Coding: DC Components (Contd..)
SIZE C d C d
DC components are differentially coded as (SIZE,Value)
The code for a SIZE is derived from the following table
Ex mpl : If DC mp t is 40
-
8/12/2019 Video Technology 2013-14 - Minor II
242/282
SIZE CodeLength
Code
0 2 00
1 3 010
2 3 011
Example: If a DC component is 40and the previous DCcomponent is 48. Thedifference is -8. Therefore itis coded as:
1010111
0111: The value for re resentin 8
242
4 3 101
5 3 110
6 4 1110
7 5 11110
8 6 111110
9 7 1111110
10 8 11111110
11 9 111111110
(see Size_and_Value table)101: The size from the same table
reads 4. The correspondingcode from the table at left is101.
Huffman Table for DC component SIZE field
-
8/12/2019 Video Technology 2013-14 - Minor II
243/282
A TV till i (f ) i I di h 720 483 i l
Why Compression?
-
8/12/2019 Video Technology 2013-14 - Minor II
244/282
A TV still image (frame) in India has 720 x 483 pixels
Hence the total amount of data for the image
= 720 x 483 x 3 bytes
= 1043280 bytes
Considering there are 25 such frames in a second, the data rate
= 1043280 x 8 x 25 bits/sec
= 208656000 bits/sec
= 208 Mbits/sec !!!
But, the cable that comes to our house can carry data rates of upto only 40Mbits/sec :-((
Compression
Store more images
-
8/12/2019 Video Technology 2013-14 - Minor II
245/282
Store more images Transmit images in less time
JPEG (Joint Photographic Experts Group) Popular standard format for representing digital images in a
compressed form
Provides for a number of different modes of operation Mode used in this class provides high compression ratios using DCT
(discrete cosine transform)
Image data divided into blocks of 8 x 8 pixels
3 steps performed on each block
DCT
Quantization
Huffman encoding
Compression
L l l ( id l d)
-
8/12/2019 Video Technology 2013-14 - Minor II
246/282
Lossless or lossy(widely used)
Color Transform (RGB to YCbCr)
Downsampling (4:2:2 to 4:2:0)
Noise reduction
Image captured
(R,G,B)
and digitised
246
Color Transform RGB to YCbCr
for Video class
C i f RGB
-
8/12/2019 Video Technology 2013-14 - Minor II
247/282
Conversion from RGB: Y=0.299(R-G) + G + 0.114(B-G)
Cb=0.564(B-Y)
Cr=0.713(R-Y)
247
The Matrix form:
0.299 0.587 0.114
0.168636 0.232932 0.064296
0.499813 0.418531 0.081282
Y R
Cb G
Cr B
Downsampling in Y-Cr-Cb space
for Video class
-
8/12/2019 Video Technology 2013-14 - Minor II
248/282
Compression
Lossless or lossy(widely used)
-
8/12/2019 Video Technology 2013-14 - Minor II
249/282
Lossless or lossy(widely used)
Noise reduction
Image Enhancement
Image captured
(grayscale)
and digitised
249
Example: JPEG Coding
YCbCr
DCTf(i, j)
8 x 8
F(u, v)
8 x 8
QuantizationFq(u, v) Steps Involved:
1. Discrete Cosine
Transform of each 8x8i l
-
8/12/2019 Video Technology 2013-14 - Minor II
250/282
Coding
QuantTables
8 x 8 8 x 8
Zi Za
Transform of each 8x8pixel arrayf(x,y) TF(u,v)
2. Quantization using atable or using a constant
3. Zig-Zag scan to exploitredundanc
250
DPCM
RLC
EntropyCoding
HeaderTables
Data
Scan 4. Differential Pulse CodeModulation(DPCM) onthe DC component andRun length Coding of theAC components
5. Entropy coding(Huffman) of the final
output
Exercise - DCT : Discrete Cosine Transform
DCT converts the information contained in a block(8x8) of pixels fromspatialdomain to thefrequencydomain.
A simple analogy: Consider a unsorted list of 12 numbers between 0 and3 > (2 3 1 2 2 0 1 1 0 1 0 0) C id t f ti f th li t
-
8/12/2019 Video Technology 2013-14 - Minor II
251/282
A simple analogy: Consider a unsorted list of 12 numbers between 0 and3 -> (2, 3, 1, 2, 2, 0, 1, 1, 0, 1, 0, 0). Consider a transformation of the listinvolving two steps: (1.) sort the list (2.) Count the frequency ofoccurrence of each of the numbers ->(????). : Through thistransformation we lost the spatial information but captured thefrequency information.
There are other transformations which retain the spatial information.
251
E.g., Fourier transform, DCT etc. Therefore allowing us to move back andforth between spatial and frequency domains.
1-D DCT: 1-D Inverese DCT:
F()a(u)2 f(n)cos(2n1)
16n 0
N1
a(0) 12
a(p) 1 p 0
f'(n)a(u)2 F()cos(2n1)
16 0
N1
Exercise - DCT : Discrete Cosine Transform
DCT converts the information contained in a block(8x8) of pixels fromspatialdomain to thefrequencydomain.
A simple analogy: Consider a unsorted list of 12 numbers between 0 and3 > (2 3 1 2 2 0 1 1 0 1 0 0) Consider a transformation of the list
-
8/12/2019 Video Technology 2013-14 - Minor II
252/282
A simple analogy: Consider a unsorted list of 12 numbers between 0 and3 -> (2, 3, 1, 2, 2, 0, 1, 1, 0, 1, 0, 0). Consider a transformation of the listinvolving two steps: (1.) sort the list (2.) Count the frequency ofoccurrence of each of the numbers ->(4,4,3,1). : Through thistransformation we lost the spatial information but captured thefrequency information.
There are other transformations which retain the spatial information.
252
E.g., Fourier transform, DCT etc. Therefore allowing us to move back andforth between spatial and frequency domains.
1-D DCT: 1-D Inverese DCT:
F()a(u)2 f(n)cos(2n1)
16n 0
N1
a(0) 12
a(p) 1 p 0
f'(n)a(u)2 F()cos(2n1)
16 0
N1
transforming the data from the spatial domain to spatial
frequency domain
Transform coding for images
-
8/12/2019 Video Technology 2013-14 - Minor II
253/282
Discrete Cosine Transform -
1-D DISCRETE COSINE TRANSFORM
DCT
1 )12()()()(
N uxfC
-
8/12/2019 Video Technology 2013-14 - Minor II
254/282
1
0 2)12(cos)()()(
N
x NuxxfuauC
,,,
1,,1
2
01
)(
NuN
uN
ua
2-D DISCRETE COSINE TRANSFORM
DCT
-
8/12/2019 Video Technology 2013-14 - Minor II
255/282
N
vy
N
uxyxfvauavuC
N
x
N
y 2
)12(cos
2
)12(cos),()()(),(
1
0
1
0
Nvy
NuxvuCvauayxf
u v 2cos
2cos),()()(),(
0 0
1,,1,0, Nvu
2-D DCT
Images are two-dimensional; How do you perform 2-D DCT? Two series of 1-D transforms result in a 2-D transform as demonstrated in
the figure below
-
8/12/2019 Video Technology 2013-14 - Minor II
256/282
1-D 1-D
j)f(i,
256
Row-wise Co umn-wise
8x8 8x8 8x8
v)F(u,
r F(0,0) is called the DC component and the rest of F(i,j) are calledAC components
-
8/12/2019 Video Technology 2013-14 - Minor II
257/282
Quality
-
8/12/2019 Video Technology 2013-14 - Minor II
258/282
258
-
8/12/2019 Video Technology 2013-14 - Minor II
259/282
259
-
8/12/2019 Video Technology 2013-14 - Minor II
260/282
DCT step
Transforms original 8 x 8 block into a cosine-frequency domain
-
8/12/2019 Video Technology 2013-14 - Minor II
261/282
Transforms original 8 x 8 block into a cosine-frequency domain Upper-left corner values represent more of the essence of the image
Lower-right corner values represent finer details Can reduce precision of these values and retain reasonable image quality
orwar ormu a C(h) = if (h == 0) then 1/sqrt(2) else 1.0 Auxiliary function used in main function F(u,v)
F(u,v) = x C(u) x C(v) x=0..7 y=0..7 fxy x cos((2u + 1)u/16) x cos((2y + 1)v/16) Gives encoded pixel at row u, column v
fxy is original pixel value at row x, column y
IDCT (Inverse DCT) Reverses process to obtain original block (not needed for this design)
What happens when DCT is
performed?
-
8/12/2019 Video Technology 2013-14 - Minor II
262/282
Example: 2D signal
-
8/12/2019 Video Technology 2013-14 - Minor II
263/282
Energy concentrated
in low-frequency
region (using DCT)
Basis of DCT
-
8/12/2019 Video Technology 2013-14 - Minor II
264/282
2-D Basis Functions N=4
0 1 2 3
u
v
-
8/12/2019 Video Technology 2013-14 - Minor II
265/282
0
1
2
3
Quantization step
Achieve high compression ratio by reducing imagequality
-
8/12/2019 Video Technology 2013-14 - Minor II
266/282
quality Reduce bit precision of encoded data
Fewer bits needed for encoding One way is to divide all values by a factor of n
Simple right shifts can do this
Dequantization would reverse process fordecompression
1150 39 -43 -10 26 -83 11 41
-81 -3 115 -73 -6 -2 22 -5
14 -11 1 -42 26 -3 17 -38
2 -61 -13 -12 36 -23 -18 5
44 13 37 -4 10 -21 7 -8
36 -11 -9 -4 20 -28 -21 14
-19 -7 21 -6 3 3 12 -21
-5 -13 -11 -17 -4 -1 7 -4
144 5 -5 -1 3 -10 1 5
-10 0 14 -9 -1 0 3 -1
2 -1 0 -5 3 0 2 -5
0 -8 -2 -2 5 -3 -2 1
6 2 5 -1 1 -3 1 -1
5 -1 -1 -1 3 -4 -3 2
-2 -1 3 -1 0 0 2 -3
-1 -2 -1 -2 -1 0 1 -1
After being decoded using DCT After quantization
Divide each cells
value by 8
Quantization
Why? -- To reduce number of bits per sample
F(u,v) = round(F(u,v)/q(u,v))
Example: 101101 = 45 (6 bits).Truncate to 4 bits: 1011 = 11. (Compare 11 x 4 =44 against 45)
-
8/12/2019 Video Technology 2013-14 - Minor II
267/282
( p g )Truncate to 3 bits: 101 = 5. (Compare 8 x 5 =40 against 45)Note, that the more bits we truncate the more precision we lose
Quantization error is the main source of the Lossy Compression.
Uniform Quantization:
267
q(u,v) is a constant.
Non-uniform Quantization -- Quantization Tables
Eye is most sensitive to low frequencies (upper left corner in frequencymatrix), less sensitive to high frequencies (lower right corner)
Custom quantization tables can be put in image/scan header.
JPEG Standard defines two default quantization tables, one each forluminance and chrominance.
Fact : Human Visual System (HVS) does not distinguish fine details
Thresholding & Quantisation
-
8/12/2019 Video Technology 2013-14 - Minor II
268/282
y ( ) gbelow a certain luminance level
Fact : HVS is less sensitive to high spatial frequency changes
Therefore,
replace values below a certain threshold (a function of
frequency) by 0
quantize the resultant values with an accuracy decreasing withincreasing spatial frequencies
The result of Thresholding and
Quantization
-
8/12/2019 Video Technology 2013-14 - Minor II
269/282
The zig-zag scan
-
8/12/2019 Video Technology 2013-14 - Minor II
270/282
(129) ; 1 ; 0 ; 0 ; 1 ; 1 ; 1 ; 1 ; 0 ; 0 ; -1 ; -1 ; 0 ; -1 ; 2 ; 1 ; -1 ; 0 ; 0 ; 0 ; 1 ; -1 ;
The resultant sequence
-
8/12/2019 Video Technology 2013-14 - Minor II
271/282
0 ; 1 ; 1 ; 0 ; 0 ; 0 ; 0 ; 0 ; -1 ; 0 ; 0 ; 0 ; 0 ; 0 ; 0 ; 1 ; 0 ; 0 ; 0 ; 0 ; 0 ; 0 ; 0 ;0 ; 0 ; 0 ; 0 ; 0 ; 0 ; 0 ; 0 ; 0 ; 0 ; 0 ; 1 ; 0 ; 0 ; 0 ; 0 ; 0 ; 0 ; 0
Huffman Coding
Huffman coding is the most popular technique
-
8/12/2019 Video Technology 2013-14 - Minor II
272/282
Huffman coding is the most popular techniquefor removing coding redundancy.
Unique prefix property
Instantaneous decodin ro ert
Optimality
JPEG(fixed, not optimal)
272
-
8/12/2019 Video Technology 2013-14 - Minor II
273/282
Huffman encoding example
Pixel frequencies on left
Pixel value 1 occurs 15 times
Pixel value 14 occurs 1 time
Build Huffman tree from bottomup
-
8/12/2019 Video Technology 2013-14 - Minor II
274/282
Create one leaf node for eachpixel value and assignfrequency as nodes value
Create an internal node byoinin an two nodes whose
23
5
6
4-1 15x
0 8x
-2 6x
-1 00
0 100
-2 110
Pixel
frequenciesHuffman tree
Huffman
codes
sum is a minimal value This sum is internal nodesvalue
Repeat until complete binarytree
Traverse tree from root to leaf toobtain binary code for leafs pixel
value Append 0 for left traversal, 1
for right traversal
Huffman encoding is reversible
No code is a prefix of anothercode
144
5 3 2
1 0 -2
-1
-10 -5 -3
-4 -8 -9614
1 1
2
1 1
2
1
22
4
3
5
4
65
9
5
1
0
5
1
15
1
4
6
17
8
181
5
9
1
x
2 5x3 5x
5 5x
-3 4x
-5 3x
-10 2x
144 1x
-9 1x
-8 1x
-4 1x
6 1x14 1x
2 1110
3 1010
5 0110
-3 11110
-5 10110
-10 01110
144 11111
-9 11111
-8 10111