Colour Image Processing

molassesitalianΤεχνίτη Νοημοσύνη και Ρομποτική

6 Νοε 2013 (πριν από 3 χρόνια και 10 μήνες)

57 εμφανίσεις

Colour Image
P
rocessing

Web reference
www.cse.msu.edu/~stockman/Book/book.html

Colour Perception


Physics of light







Human Perception


Land Colour Mondrains

Human Vs Hardware

Hardware



Colour Cameras


Mosaic


3 chip


Frame grabbers


3 frame buffers


Red, Green and Blue

Image Physics


Colour Depends on


Spectral reflectance of surface


Spectrum of illumination


Spectral response of sensors


Hue, Saturation, Intensity (HSI)


Intensity


Hue (light of a particular wavelength)


Saturation (degree of dominance of a colour)

CIE standards for colour
reproduction


CIE XYZ, CIE xyY, CIE L*u*v*, CIE
L*a*b*,…


Colour Constancy


Illumination independent recognition


Match colours under varying illumination


Land Mondrian

Hue, Saturation, Intensity



S

H

Red

Yellow

Green

Cyan

Blue

Magenta













B
G
R
B
G
R
S
B
G
B
R
G
R
B
G
R
H
B
G
R
I























,
,
min
3
1
2
2
cos
3
1
2
1
Other colour spaces


Opponent




YIQ (NTSC)






































B
G
R
black
white
yellow
blue
green
red
1
1
1
2
1
1
1
2
1



































B
G
R
Q
I
Y
31
.
0
52
.
0
21
.
0
32
.
0
28
.
0
6
.
0
11
.
0
59
.
0
3
.
0
Colour Vision


Why?


Feature tollerant to


Scale


Optical distortion


View point


A natural cue


Useful in addition to geometric features


But?


May not be intrinsic (can lepard change its spots)


Objects contain many colours

Colour Image Processing


Pixel by pixel classification is error prone


Noise


Specular reflections


Hue unreliable when saturation is low


Saturation unreliable when intensity is low


Colour Edges


Better quality edges than intensity alone


Extra computation


Fusion ?


Not many new edges


Colour Histograms


Histograms tolerant to


Translation, rotation, scale and partial occlusion


Image Database retrieval


Swain and Ballard 1991


Create colour histogram of images


Match histograms to retrieve images


Find similar images

Colour Histograms


Reduce the complexity


2
6
, 2
4
,…


Concatenate separate RGB histograms into one


Intersection of h(i) and h(m) min over all K bins


Match can normalise over those bins defined in
the model


This removes the contribution of background pixels in
h(i)



Other metrics possible


Examples:

Back Projection


Locate a region within an image containing
a learned object


Remove intensity

component


Smooth Histograms


Colour Profile


Characterize flaws


Recognize flaw signature

Colour Profiles


Histogram


Good


Flaws


Profile


Colour unique to
flaws


Classify based on
unique colours

Face Detection With Colour


Human Skin tones lie within narrow range


Face recognition


Image filters for porn on the web


Other objects also have similar colour


Colour Segmentation followed by


Connected component analysis


Morphology


Blob analysis


Multi
-
spectral imaging


IR, X
-
ray, Radar, MRI…..


GIS systems


Medical systems


Pseudo Colour (Thematic) Images


Colour placed on images to communicate
information


Doppler information on ultrasound images


Depth information on GIS images