Date post: | 04-Jun-2018 |
Category: |
Documents |
Upload: | doannguyet |
View: | 222 times |
Download: | 0 times |
Interactive Systems Laboratories, Universität Karlsr uhe (TH)Dr. Edgar Seemann, 23.10.09 1
Visuelle Perzeption für Mensch-Maschine Schnittstellen
Vorlesung, WS 2009
Prof. Dr. Rainer StiefelhagenDr. Edgar Seemann
Institut für AnthropomatikUniversität Karlsruhe (TH)
http://[email protected]
Dr. Edgar Seemann, 23.10.09 2
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciTermine (1)
2
TBA21.12.2009
Stereo and Optical Flow18.12.2009
Termine Thema19.10.2009 Introduction, Applications
23.10.2009 Basics: Cameras, Transformations, Color
26.10.2009 Basics: Image Processing
30.10.2009 Basics: Pattern recognition
02.11.2009 Computer Vision: Tasks, Challenges, Learning, Performance measures
06.11.2009 Face Detection I: Color, Edges (Birchfield)
09.11.2009 Project 1: Intro + Programming tips13.11.2009 Face Detection II: ANNs, SVM, Viola & Jones
16.11.2009 Project 1: Questions
20.11.2009 Face Recognition I: Traditional Approaches, Eigenfaces, Fisherfaces, EBGM
23.12.2009 Face Recognition II
27.11.2009 Head Pose Estimation: Model-based, NN, Texture Mapping, Focus of Attention
30.11.2009 People Detection I
03.12.2009 People Detection II
07.12.2009 Project 1: Student Presentations, Project 2: Intro11.12.2009 People Detection III (Part-Based Models)
14.12.2009 Scene Context and Geometry
Interactive Systems Laboratories, Universität Karlsr uhe (TH)Dr. Edgar Seemann, 23.10.09 3
Book Recommendations
� „Computer Vision a Modern Approach“, David Forsyth and Jean Ponce
� „Pattern Classification“, Duda&Hart
� „Multiple View Geometry“, Hartley&Zisserman
Interactive Systems Laboratories, Universität Karlsr uhe (TH)Dr. Edgar Seemann, 23.10.09 4
Organisatorisches
� Programmier-Aufgaben� Gruppen a 3 Studenten
� Ergebnisse haben Auswirkung auf Gesamtnote
� C++, Okapi � Mehr Details und eine kleine Einführung am 9. November
� Bitte tragt euch in die Liste ein
Interactive Systems Laboratories, Universität Karlsr uhe (TH)Dr. Edgar Seemann, 23.10.09 6
Programmier-Umgebung
� Ihr könnt mit eurem Lieblings-Editor programmieren
� Empfehlung:� Eclipse (CDT)
� QtCreator
� vi, emacs ;-)
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ci
Basics:
Cameras, Transformations and Color Spaces
WS 2009/10
Dr. Edgar Seemann
Dr. Edgar Seemann, 23.10.09 8
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciToday‘s Overview
� Cameras and� Pinhole Camera Model
� Image formation� perspective projection
� calibration
� Transformations� Euclidian, similarity, affine, projective transformation
� some invariants
� ColorCredits: Bernt Schiele (Darmstadt), Bastian Leibe (Aachen),Kirsten Grauman (University of Texas)David Forsyth (University of Illinois)James Rehg (Georgia Institute of Technology)
Dr. Edgar Seemann, 23.10.09 9
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciQuiz
� When was photography invented?
� By whom? � Exposure time?
http://www.digicamhistory.com
Dr. Edgar Seemann, 23.10.09 10
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciQuiz
� When was photography invented? 1825
� By whom? Niépce� Heliography
� Exposure time? 8 hours
� William Henry Fox Talbot inventsthe calotype in 1834 which prettymuch invents the negative
Dr. Edgar Seemann, 23.10.09 11
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciFirst production camera?
Dr. Edgar Seemann, 23.10.09 12
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciFirst production camera?
� 1839. Daguerrotype
Dr. Edgar Seemann, 23.10.09 13
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciBeginning of hobby photography?
� 1900 Kodak Brownie
Dr. Edgar Seemann, 23.10.09 14
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciQuiz
� Who did the first color photography?
� When?
Dr. Edgar Seemann, 23.10.09 15
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciQuiz
� Who did the first color photography? � James Cleark Maxwell (also known
from the Maxwell equations) When? 1861
� Oldest color photos still preserved: Prokudin-Gorskiihttp://www.loc.gov/exhibits/empire/
Dr. Edgar Seemann, 23.10.09 16
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciProkudin-Gorskii
� Digital restoration
Dr. Edgar Seemann, 23.10.09 17
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciProkudin-Gorskii
Dr. Edgar Seemann, 23.10.09 18
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciProkudin-Gorskii
Dr. Edgar Seemann, 23.10.09 21
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciInstant photography?
Dr. Edgar Seemann, 23.10.09 22
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciInstant photography?
� 1947, Edwin Land (Polaroid founder)
Dr. Edgar Seemann, 23.10.09 39
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciFirst microprocessor in a camera
� Canon AE-11976
Dr. Edgar Seemann, 23.10.09 57
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciFirst digital camera?
Dr. Edgar Seemann, 23.10.09 58
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciFirst digital camera?
� 1975, Steve Sasson, Kodak
� Uses ccd from Fairchild semiconductor, A/D from Motorola, .01 megapixels, 23 second exposure, recorded on digital cassette
Dr. Edgar Seemann, 23.10.09 64
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciFirst megapixel sensor
� Of reasonable size?
Dr. Edgar Seemann, 23.10.09 65
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciFirst megapixel sensor
� Of reasonable size?
� (Kodak) Videk 1987, 1.4MPixels
Dr. Edgar Seemann, 23.10.09 75
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciReferences
� http://www.digicamhistory.com/
� http://www.photo.net/history/timeline� http://inventors.about.com/library/inventors/blphotography.htm� http://www.loc.gov/exhibits/empire/� http://www.spartacus.schoolnet.co.uk/USAphotographers.htm� http://www.eyeconart.net/history/photography.htm� http://www.scphoto.com/html/history.html� http://www.g4tv.com/callforhelparchive/features/44534/Witness_to_History_
The_Digital_Camera.html� http://www.digicamhistory.com/� http://www-users.mat.uni.torun.pl/~olka/� http://inventors.about.com/od/pstartinventions/a/Photography.htm� http://www.ted.photographer.org.uk/camera_designs_3.htm� http://accad.osu.edu/~waynec/history/timeline.html
� http://en.wikipedia.org/wiki/History_of_the_single-lens_reflex_camera
Dr. Edgar Seemann, 23.10.09 77
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciCameras
� First photograph due to Niépce(1820s)
� First digital photographs (Bell, Texas Instruments, Kodak 1970s)
� Basic abstraction is the pinhole camera� lenses required to ensure image is not too dark
� various other abstractions can be applied
Dr. Edgar Seemann, 23.10.09 78
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciPinhole Camera
� (simple) standard and abstract model today� box with a small hole in it
� Pinhole cameras work in practice
Dr. Edgar Seemann, 23.10.09 79
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciDiffraction limit
� Optimal size for visible light: sqrt(f)/28 (in millimiters) where f is focal length
Dr. Edgar Seemann, 23.10.09 80
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciPinhole too big-many directions areaveraged, blurring theimage
Pinhole too small-diffraction effects blurthe image
Generally, pinhole cameras are dark, becausea very small set of raysfrom a particular pointhits the screen.
Dr. Edgar Seemann, 23.10.09 81
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciProperties
� Distant objects are smaller
� Moving the image plane simply scales the image
� Each set of parallel lines (=direction) meets at a different point� The vanishing point for this direction
� Sets of parallel lines on the same plane lead to collinear vanishing points. � The line is called thehorizon for that plane
� Good ways to spot faked images� scale and perspective don’t work
� vanishing points behave badly
� Shadows and ligthing are not consistent
Dr. Edgar Seemann, 23.10.09 82
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciPinhole Camera Model
� origin of the coordinate system is O
� the optical axis of the camera is parallel to the z-axis of the coordinate system
� f is the focal length
� the z-coordinate of the image plane is: -f
image planexy
z
f
O
Dr. Edgar Seemann, 23.10.09 83
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciPinhole Camera Model
� the image of an object is the projection of the object onto the image plane
� intersection of all projection-lines is the focal point F of the camera� the origin is defined to be the focal point of the camera: F = O
f
xy
z
image plane
F=O
Dr. Edgar Seemann, 23.10.09 84
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciPinhole Camera Model
� due to similar triangles and the intercept theorems:
′z = − f′x
− f=
x
z ⇒ ′x =
− fx
z
′y
− f=
y
z ⇒ ′y =
− fy
z
xy
z
f P
′P
P =x
y
z
′P =
′x
′y
′z
Dr. Edgar Seemann, 23.10.09 85
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciMathematical Simplification
� translation of the image plane in front of the focal point
′z = f
′x =fx
z
′y =fy
z
xy
z
f P
′P
image plane
Dr. Edgar Seemann, 23.10.09 86
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciLinear Transformation
Goal: formulation of the projection as a linear transformation
- i.e. find matrix A such that:
′P =′x
′y
′z
= A
x
y
z
= A P
• Transformation matrix depends on z
• It should therefore be different for each point p
-> NOT WORKING
′z = f
′x =fx
z
′y =fy
z
Dr. Edgar Seemann, 23.10.09 87
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciHomogenous Coordinates
� define equivalence class:
′x
′y
′z
→
′x
′y
′z
1
≅
s ′x
s ′y
s ′z
s
=
sfx
z
sfy
zsf
s
=
fx
fy
fz
z
x
y
z
→
x
y
z
1
≅
sx
sy
sz
s
≅
Dr. Edgar Seemann, 23.10.09 88
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciMathematical Simplification
′x
′y
′z
→
′x
′y
′z
1
≅
fx
fy
fz
z
=
f 0 0 0
0 f 0 0
0 0 f 0
0 0 1 0
x
y
z
1
fx
fy
fz
z
= A
x
y
z
1
Dr. Edgar Seemann, 23.10.09 89
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciCompletion of the Projection
� so far:� projection of a point P onto the image plane
� what’s missing:� pixel-coordinates (u,v) of the projected points
v = − k v ′y + v 0
u = k u ′x + u 0
′x
′y
u
v
u 0
v 0
image plane
With ku and kv scaling factorswhich denote the ratio between world and pixel coordinates
Dr. Edgar Seemann, 23.10.09 90
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciCompletion of the Projection
� in matrix notation:
� again in homogeneous coordinates:
u
v
=
k u 0
0 − k v
′x
′y
+
u 0
v 0
u
v
1
=k u 0 0
0 − k v 0
0 0 0
u 0
v 0
1
′x
′y
′z
1
Dr. Edgar Seemann, 23.10.09 91
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciComplete Projection
� multiplication of the two projection matrices:
u
v
1
=k u 0 0
0 − k v 0
0 0 0
u 0
v 0
1
f 0 0 0
0 f 0 0
0 0 f 0
0 0 1 0
x
y
z
1
u
v
1
=k u f 0 u 0
0 − k v f v 0
0 0 1
0
0
0
x
y
z
1
Dr. Edgar Seemann, 23.10.09 92
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciPerspective Projection
� 4 parameters:
� have to be known to perform the projection
� called ‘internal camera parameters’ since they depend on the camera only
� perform calibration to estimate those parameters
α u uk f=α v vk f= −u 0
v 0
Dr. Edgar Seemann, 23.10.09 93
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciCalibration
� calibration object� point-coordinates known in 3D
� left: model of the calibration object
� right: image taken with the camera
Dr. Edgar Seemann, 23.10.09 94
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciCalibration
� left: edge detection
� right: extraction of the corner-points of the squares
Dr. Edgar Seemann, 23.10.09 95
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciCalibration
� transform the (corner-)points pi of the calibration object into the coordinate system of the camera: � R: rotation (3 degrees of freedom), 3x3 matrix� t: translation (3 degrees of freedom), 3x1 vector
� in homogenous coordinates:
x i
y i
z i
= R p i + t
x i
y i
z i
1
=R t
0 1
p i
1
=
r1 1 r1 2 r13 t x
r2 1 r22 r2 3 t y
r3 1 r32 r3 3 t z
0 0 0 1
p i
1
Dr. Edgar Seemann, 23.10.09 96
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciCalibration
� the observation of the projected points are in image-coordinates
� i.e. we have to project the pi onto the image plane:
u i
v i
1
=α u 0 u 0
0 α v v 0
0 0 1
0
0
0
x i
y i
z i
1
x i
y i
z i
1
=R t
0 1
p i
1
=
r1
r 2
r 3
0
t x
t y
t z
1
p i
1
Dr. Edgar Seemann, 23.10.09 97
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciCalibration
� the resulting matrix is a 3x4 matrix
� theoretically this matrix has 12 parameters,
� in reality it has only 10 degrees of freedom
� therefore we need (minimally) 5 points to solve the system of linear equations (more points are better…)
su i
sv i
s
=α u 0 u 0
0 α v v 0
0 0 1
0
0
0
r1
r 2
r 3
0
t x
t y
t z
1
p i
1
su i
sv i
s
=α u r1 + u 0 r 3
α v r 2 + v 0 r 3
r 3
α u t x + u 0 t z
α v t y + v 0 t z
t z
p i
1
Dr. Edgar Seemann, 23.10.09 98
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciInternal and External Parameters
� internal parameters of the camera:
� external parameters of the cameras:� rotation R
� translation t
α u uk f=α v vk f= −u 0
v 0
Dr. Edgar Seemann, 23.10.09 99
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciThe reason for lenses
Dr. Edgar Seemann, 23.10.09 100
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciHow to trace rays
� Start by rays through the center
Dr. Edgar Seemann, 23.10.09 101
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciHow to trace rays
� Start by rays through the center
� Choose focal length, trace parallels
f
Dr. Edgar Seemann, 23.10.09 102
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciHow to trace rays
� Start by rays through the center
� Choose focal length, trace parallels
� You get the focus plane for a given scene plane� All rays coming from points on a plane parallel to the lens are focused on
another plane parallel to the lens
f
Dr. Edgar Seemann, 23.10.09 103
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciFocusing
� To focus closer than infinity� Move the sensor/film further than the focal length
f
Dr. Edgar Seemann, 23.10.09 104
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciThin lens formula
fDD’
Dr. Edgar Seemann, 23.10.09 105
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciThin lens formula
fDD’
Dr. Edgar Seemann, 23.10.09 106
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciThin lens formula
fDD’
Similar triangles everywhere!
y’
y
y’/y = D’/D
Dr. Edgar Seemann, 23.10.09 107
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciThin lens formula
fDD’
Similar triangles everywhere!
y’
y
y’/y = D’/D
y’/y = (D’-f)/f
Dr. Edgar Seemann, 23.10.09 108
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciThin lens formula
fDD’
1D’ D
1 1f
+ =
Dr. Edgar Seemann, 23.10.09 109
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciMinimum focusing distance
� By symmetry, an object at the focal length requires the film/sensor to be at infinity.
Film/sensor
Rays from infinity
Rays from object at f
Dr. Edgar Seemann, 23.10.09 111
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciField of view & focusing
� What happens to the field of view when one focuses closer?� It's reduced
film focused at infinity
film focused close
Dr. Edgar Seemann, 23.10.09 112
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciFocal length in practice
24mm
50mm
135mm
Dr. Edgar Seemann, 23.10.09 120
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciRecap
� Pinhole is the simplest model of image formation
� Lenses gather more light� But get only one plane focused
� Focus by moving sensor/film
� Cannot focus infinitely close
� Focal length determines field of view� From wide angle to telephoto
� Depends on sensor size
Dr. Edgar Seemann, 23.10.09 121
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciSpherical aberration
Dr. Edgar Seemann, 23.10.09 122
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciLens systems
Dr. Edgar Seemann, 23.10.09 123
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciVignetting
Dr. Edgar Seemann, 23.10.09 124
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciOther (possibly annoying) phenomena
� Chromatic aberration� Light at different wavelengths follows different paths; hence, some
wavelengths are defocussed� Machines: coat the lens� Humans: live with it
� Scattering at the lens surface� Some light entering the lens system is reflected off each surface it
encounters (Fresnel’s law gives details)� Machines: coat the lens, interior� Humans: live with it (various scattering phenomena are visible in
the human eye)
� Geometric phenomena (Barrel distortion, etc.)
Dr. Edgar Seemann, 23.10.09 125
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciA simpler abstraction model
� perspective projection can be approximated by a ‘scaled orthographic’ projection� if the object is far away from the camera (w.r.t. the focal length of
the camera
� then we can neglect the depth of the object (the object is regarded as small and flat)
� (analogous: the focal length of the camera is infinite)
� orthographic projection:� the projection is orthogonal to the image plane
xyP′P
Dr. Edgar Seemann, 23.10.09 126
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciScaled Orthographic Projection
� Reason:� if the object is ‘nearly’ planar, then all points on the object have
‘nearly’ the same distance to the camera� the scale-factor of the projection is therefore ‘nearly’ constant for
all points on the object
� this is called the ‘scaled orthographic’ projection� this is quite often a reasonable assumption
′ = =xfx
zs x
00′ =x
fx
z
′ =yfy
z′ = =y
fy
zs y
00
Dr. Edgar Seemann, 23.10.09 127
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ci
2D Alignment
Dr. Edgar Seemann, 23.10.09 128
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ci
� 2 objects in the database
� new test image - question: which object
Recognition: Example
Dr. Edgar Seemann, 23.10.09 129
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciRecognition: Example
� Alignment:� alignment: transform the image such that one object is
in complete correspondence with the other object
� Objects are assumed to be segmented (i.e. we know which pixels belong to the background and which to the object)
� estimate the parameters of the transformation
Dr. Edgar Seemann, 23.10.09 130
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciRecognition: Example
� translation:
� rotation:
� scaling and mirroring
� what remains is to compare the 2 (segmented) binary object-masks
Dr. Edgar Seemann, 23.10.09 131
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciDegrees of Freedom (DoF) for Objects
� Euclidean transformation� transformations in the image plane� rotation (1 DoF)� translation (2 DoF)
� invariants under these transformations� length - yes� angle - yes� length-ratio - yes� parallelism - yes
� object can be localized with 2 points� 2 points in the image plane have 4 coordinates - which can be used to
determine the 3 DoF
Dr. Edgar Seemann, 23.10.09 132
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciDegrees of Freedom (DoF) for Objects
� Similarity transformation:� rotation (1 DoF)� translation (2 DoF)� scaling of the object (1 DoF)
� invariants under these transformations� length - no� angle - yes� length-ratio - yes� parallelism - yes
� object can be localized with 2 points� 2 points in the image plane have 4 coordinates - which can be used to
determine the 4 DoF
Dr. Edgar Seemann, 23.10.09 133
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciDegrees of Freedom (DoF) for Objects
� Affine Transformation:� has 6 DoF
� 2x2 = 4 parameters + 2 parameters for translation
� rotation(1), translation (2), scaling (1)
� the remaining 2 DoF are � non-uniform scaling of the axes (1 DoF)
� shear of the axes (1 DoF)
Dr. Edgar Seemann, 23.10.09 134
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciDegrees of Freedom (DoF) for Objects
� invariants under affine transformation:� length- no
� angle - no (because of shear)
� length-ratio - no (because of non-uniform scaling)
� parallelism - yes
� object can be localized with 3 points� 3 points in the image plane have 6 coordinates -which
can be used to determine the 6 DoF
Dr. Edgar Seemann, 23.10.09 135
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciDegrees of Freedom (DoF) for Objects
� Perspective transformation� adds 2 DoF
� Scaling depending on x-, y-position
� invariants under projective transformation:� length- no
� angle - no
� length-ratio - no
� parallelism - no
� object can be localizedwith 4 points:� 4 points need to determine
the complete coordinate system…
Dr. Edgar Seemann, 23.10.09 151
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ci
Color
Dr. Edgar Seemann, 23.10.09 152
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciOutline
� Color and Radiometry
� Human Color Perception
� Color spaces
� Lightness and Color Constancy
� Physics-based Vision: Specularities
Dr. Edgar Seemann, 23.10.09 153
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciWhy does a visual system need color?
(an incomplete list…)� To tell what food is edible.
� To distinguish material changes from shading changes.
� To group parts of one object together in a scene.
� To find people’s skin.
� Check whether a person’s appearance looks normal/healthy.
� To compress images
Dr. Edgar Seemann, 23.10.09 154
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciColor and Radiometry
� What is color ?
Dr. Edgar Seemann, 23.10.09 155
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciWhat is Color?
� A perceptual attribute of objects and scenes constructed by the visual system
� A quantity related to the wavelength of light in the visible spectrum
� A challenge� “There are no second-rate brains in color vision”
– Edwin Land
Dr. Edgar Seemann, 23.10.09 156
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
cicolor
Dr. Edgar Seemann, 23.10.09 158
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciReflectance Model
Dr. Edgar Seemann, 23.10.09 160
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciIllumination Spectra
Blue sky Tungsten light bulb
Dr. Edgar Seemann, 23.10.09 161
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciReflectance Spectra
Spectral albedoes for several different leaves, with color names attached. Notice that different colourstypically have different spectral albedo, but that different spectral albedoes may result in the same perceived color (compare the two whites). Spectral albedoes are typically quite smooth functions. Measurements by E.Koivisto.
Dr. Edgar Seemann, 23.10.09 162
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciColor names for cartoon spectra
400 500 600 700 nm
400 500 600 700 nm
400 500 600 700 nm
red
gree
nbl
ue
400 500 600 700 nm
cyan
mag
enta
yello
w
400 500 600 700 nm
400 500 600 700 nm
Dr. Edgar Seemann, 23.10.09 163
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciAdditive color mixing
400 500 600 700 nm
400 500 600 700 nm
red
gree
n
Red and green make…
400 500 600 700 nm
yello
w
Yellow!
When colors combine by adding the color spectra. Example color displays that follow this mixing rule: CRT phosphors, multiple projectors aimed at a screen
Dr. Edgar Seemann, 23.10.09 164
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciSubtractive color mixing
When colors combine by multiplying the color spectra. Examples that follow this mixing rule: most photographic films, paint, cascaded optical filters, crayons.
400 500 600 700 nm
cyan
yello
w
400 500 600 700 nm
Cyan and yellow (in crayons,called “blue” and yellow) make…
400 500 600 700 nmGreen!gr
een
Dr. Edgar Seemann, 23.10.09 165
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciWhy specify color numerically?
� Accurate color reproduction is commercially valuable � Many products are identified by
color (“golden” arches);
� Few color names are widely recognized by English speakers -� About 10; other languages have
fewer/more, but not many more.
� It’s common to disagree on appropriate color names.
� Color reproduction problems increased by prevalence of digital imaging - eg. digital libraries of art. � How do we ensure that
everyone sees the same color?
Dr. Edgar Seemann, 23.10.09 166
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciAn assumption that sneaks in here
� For now we will assume that the spectrum of the light arriving at your eye completely determines the perceived color.
� But we know color appearance really depends on:� The illumination
� Your eye’s adaptation level
� The colors and scene interpretation surrounding the observed color.
Dr. Edgar Seemann, 23.10.09 167
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciColor Matching Process
Basis for industrial color standards
Dr. Edgar Seemann, 23.10.09 168
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciColor Matching Experiment 1
Image courtesy Bill Freeman
Dr. Edgar Seemann, 23.10.09 169
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciColor Matching Experiment 1
Image courtesy Bill Freeman
Dr. Edgar Seemann, 23.10.09 170
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciColor Matching Experiment 1
Image courtesy Bill Freeman
Dr. Edgar Seemann, 23.10.09 171
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciColor Matching Experiment 1
Image courtesy Bill Freeman
Dr. Edgar Seemann, 23.10.09 172
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciColor Matching Experiment 2
Image courtesy Bill Freeman
Dr. Edgar Seemann, 23.10.09 173
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciColor Matching Experiment 2
Image courtesy Bill Freeman
Dr. Edgar Seemann, 23.10.09 174
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciColor Matching Experiment 2
Image courtesy Bill Freeman
Dr. Edgar Seemann, 23.10.09 175
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciColor Matching Experiment 2
Image courtesy Bill Freeman
Dr. Edgar Seemann, 23.10.09 176
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciConclusion from Color Matching
� Three primaries are sufficient for most people to reproduce arbitrary colors� The human eye normally contains only three types of
color receptors, called cone cells� Each color receptor responds to different ranges of the
color spectrum. � Humans respond to the light stimulus via a three-
dimensional sensation, which generally can be modeled as a mixture of three primary colors.
Dr. Edgar Seemann, 23.10.09 177
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciCaveat: Context Matters !
Figure courtesy ofD. Forsyth
Dr. Edgar Seemann, 23.10.09 180
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciPrinciple of Trichromaticity
∑
∑
=
=
=
===
=
3
1
321
1
enough are primaries 3 imply that sexperiment matchingColor
(B) nm 44444 (G), nm 32526 (R), nm 16645
:Example weights.are primaries, are
:generalIn
iii
ii
m
iii
PwT
.P.P.P
wP
PwT
Dr. Edgar Seemann, 23.10.09 181
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciGrassman’s Laws
� For color matches:� symmetry: U=V <=>V=U� transitivity: U=V and V=W => U=W� proportionality: U=V <=> tU=tV� additivity: if any two (or more) of the statements
U=V, W=X, (U+W)=(V+X) are true, then so is the third
� These statements are as true as any biological law. They mean that additive color matching is linear.
Forsyth & Ponce
Dr. Edgar Seemann, 23.10.09 182
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciMeasure color by color-matching paradigm
� Pick a set of 3 primary color lights.� Find the amounts of each primary, e1, e2, e3, needed to match
some spectral signal, t.� Those amounts, e1, e2, e3, describe the color of t. If you have
some other spectral signal, s, and s matches t perceptually, then e1, e2, e3 will also match s, by Grassman’s laws.
� Why this is useful—it lets us:� Predict the color of a new spectral signal� Translate to representations using other primary lights.
Dr. Edgar Seemann, 23.10.09 184
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciHow to compute the color match for any color
signal for any set of primary colors
� Pick a set of primaries,
� Measure the amount of each primary, needed to match a monochromatic light, at each spectral wavelength (pick some spectral step size). These are called the color matching functions.
)(),(),( 321 λλλ ppp
)(),(),( 321 λλλ ccc
)(λtλ
Dr. Edgar Seemann, 23.10.09 185
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciColor matching functions for a particular set of
monochromatic primariesp1 = 645.2 nmp2 = 525.3 nmp3 = 444.4 nm
Foundations of Vision, by Brian Wandell, Sinauer Assoc., 1995
Dr. Edgar Seemann, 23.10.09 193
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciColor Spaces
� Use color matching functions to define a coordinate system for color.
� Each color can be assigned a triple of coordinates with respect to some color space (e.g. RGB).
� Devices (monitors, printers, projectors) and computers can communicate colors precisely.
Dr. Edgar Seemann, 23.10.09 202
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciSuppose you use one set of primaries and I
use another?
� We address this in 2 ways:� Learn how to translate between primaries
� Standardize on a few sets of favored primaries.
Dr. Edgar Seemann, 23.10.09 203
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciNTSC color components: Y, I, Q
−−−=
B
G
R
Q
I
Y
312.0523.0211.0
322.0274.0596.0
114.0587.0299.0
� Mathematically, color space conversions of linear color spaces are basis transformations
� Basis transformations can be expressed in terms of a matrix multiplication
Dr. Edgar Seemann, 23.10.09 204
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciNTSC - RGB
Dr. Edgar Seemann, 23.10.09 205
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciHue Saturation Value
� Attempt to describe the colors’ perceptual relationship
� Related to RGB
� Value: from black to white
� Hue: dominant color (red, orange, etc)
� Saturation: from gray to vivid color
� Exampes:RGB HSV (1, 0, 0) (0°, 1, 1) red(0, 0, 0.5) (240°, 1, 0.5) dark blue
Dr. Edgar Seemann, 23.10.09 206
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciRGB <> HSV
� Converting RGB values to HSV
Dr. Edgar Seemann, 23.10.09 211
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciLightness and Color Constancy
Dr. Edgar Seemann, 23.10.09 213
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciHuman Color Constancy
� Color constancy: hue and saturation
� Lightness constancy: gray-level
� Humans can perceive� Color a surface would have under white light (surface
color)
� Color of reflected light (separate surface color from measured color)
� Color of illuminant (limited)
Dr. Edgar Seemann, 23.10.09 214
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciLand’s Mondrian Experiments
� Squares of color with the same color radiance yield very different color perceptions
Photometer: 1.0, 0.3, 0.3 Photometer: 1.0, 0.3, 0.3
Audience: “Red” Audience: “Blue”White light Colored light
Blue
Red
Blue
Red
Dr. Edgar Seemann, 23.10.09 215
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciSpecularities
Real scene:
Figure courtesy ofSing Bing Kang
Removed specularities:
Dr. Edgar Seemann, 23.10.09 221
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciSummary Color
� Color is not that straight forward as one mighthave guessed
� Color is a significant industry with conferences, standards bodies, etc
Dr. Edgar Seemann, 23.10.09 222
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciSelected Bibliography
Vision and Art : The Biology of Seeingby Margaret Livingstone, David H. HubelHarry N Abrams; ISBN: 0810904063 208 pages (May 2002)
Vision Science
by Stephen E. PalmerMIT Press; ISBN: 0262161834760 pages (May 7, 1999)
Billmeyer and Saltzman's Principles of Color Technology, 3rd Editionby Roy S. Berns, Fred W. Billmeyer, Max SaltzmanWiley-Interscience; ISBN: 047119459X304 pages 3 edition (March 31, 2000)
Dr. Edgar Seemann, 23.10.09 224
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ciOther color references
� Reading: � Chapter 6, Forsyth & Ponce
� Chapter 4 of Wandell, Foundations of Vision, Sinauer, 1995 has a good treatment of this.
Credits: Bill Freeman
Dr. Edgar Seemann, 23.10.09 225
Com
pute
r V
isio
n fo
r H
uman
-Com
pute
r In
tera
ctio
n
Res
earc
h G
roup
, Uni
vers
itätK
arls
ruhe
(T
H)
cv:h
ci
End of Lecture