1

CS332VisualProcessingDepartmentofComputerScienceWellesleyCollege

CS332VisualProcessinginComputerandBiologicalVisionSystems

AnalysisofColor

2

Content-basedimageretrieval

2

3

Applicationsofcolorimagesegmentation

medicine,surveillance,inspection,recognition,…

4

Colorconstancy

Colorofasurfacelooks(roughly)thesameunderdifferentilluminations…

http://www.echalk.co.uk/amusements/OpticalIllusions/illusions.htm

…althoughcolorscanbeinfluencedbycontext

3

5

Surfacereflectance

R(x,y,λ)Surfacereflectancevarieswithpositionandwavelength

6

Illumination

Evenwithuniformsceneillumination,castshadowsandreflectionoflightfromothersurfacescreatesspatiallynon-uniformsurfaceillumination

E(x,y,λ)

4

7

Surfacereflectancemeetsilluminant

E(x,y,λ)

R(x,y,λ)

L(x,y,λ)=

E(x,y,l)*R(x,y,l)

8

“TheDress”– whatcolorisit?

white-gold?(30%) blue-black?(57%)???Lafer-Sousa,Hermann,Conway(2015)

5

9

Itdependsontheilluminant...

coolilluminant(bluesky)

warmilluminant(incandescent)

10

Measuringcolorbyretinalcones

AbsorptionspectraforS(short),M(medium),L(long)wavelengthcones

6

11

Luminancemeetsthecones

L(x,y,λ)

S(λ)M(λ)L(λ)

IS(x,y)=∫L(x,y,λ)S(λ)dλ

IM(x,y)=∫L(x,y,λ)M(λ)dλ

IL(x,y)=∫L(x,y,λ)L(λ)dλ

12

Land’scolor“Mondrian”experiments

PietMondrianCompositionA, 1923EdwinLand

7

13

Land’sRetinex theory*

L(x,y)=E(x,y)*R(x,y)

L(x,y):luminance

E(x,y):illuminant

R(x,y):surfacereflectance

Goal: recoversurfacereflectance

*ignorecolorfornow…

14

Land’sRetinex theory(1D)

R(x)

E(x)

L(x)

log(L(x))

derivativeoflog(L(x))

computedR(x)

8

15

Retinex theoryfor2Dcoloranalysis2Dextensions:• Land&McCann:multiple1Dpaths• Horn:2DanalysisbasedonLaplacianÑ2L• Jobson,Rahman&Woodell:appliedtoimageenhancement

Color:• PerformsameanalysisforIS(x,y),IM(x,y),IL(x,y)• TripletofvaluesRS(x,y),RM(x,y),RL(x,y)Þ color

Restrictions(Assumptions):• Reflectancechangesabruptly,illuminationchangesslowly•Matte(Lambertian)reflectancecharacteristics

16

Robson,Rahman&WoodellMorel,Petro&Sbert

Imageenhancement,Retinex style