54
IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland http://spectral.joensuu.fi/

IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Embed Size (px)

Citation preview

Page 1: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

IPCV ‘06August 21 – September 1

Budapest

Jussi ParkkinenMarkku Hauta-Kasari

Department of Computer ScienceUniversity of Joensuu, Finland

http://spectral.joensuu.fi/

Page 2: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland
Page 3: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland
Page 4: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Introduction to spectral color

Page 5: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

COLOR ANALYSIS

Page 6: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland
Page 7: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland
Page 8: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland
Page 9: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland
Page 10: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Eriväristen lehtien spektrejä

Page 11: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Spectral images from real and artificial indoor plants Kanae Miyazawa

Page 12: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Display characterictics

Page 13: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Principle of color detection

Light source

Colored object

Detection system•biological•artificial

Page 14: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Color image formation in human eye

Page 15: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

CIE Illum inations D65, A and C

0

50

100

150

200

250

300

300 350 400 450 500 550 600 650 700 750 800

Wavelength (nm)

D65

A

C

Page 16: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland
Page 17: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland
Page 18: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland
Page 19: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland
Page 20: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland
Page 21: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland
Page 22: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland
Page 23: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

RGB vs. Spectrum

• Is the spectrum needed?

• RGB is just a 3D projection of a spectrum

• RGB can produce nice colors on display, but not correct colors

Page 24: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland
Page 25: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland
Page 26: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland
Page 27: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland
Page 28: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Spectral approach to color

• In spectral approach, color is represented by color signal. This causes the color sensation

• The signal is part of electromagnetic spectrum

- in human color vision the range is 380-780 nm

• In spectral approach, we are not limited into this human visual range

Page 29: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

What should be the spectralresolution, i.e.

the sampling rate in wavelenths?

Page 30: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Spectral dependence on sampling

Page 31: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Color dependence on sampling

Page 32: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Change of RGB-values due to sampling

Page 33: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Can we reproduce a spectrum from the RGB-values?

Page 34: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Data• 1494 color samples in 200 color images

– Munsell colors, Color checker, natural colors, wall paint

– cameras: Fuji FinePix and Canon Powershot

– illuminants: A and D65

Page 35: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Methods

1. Polynomial model (R, G, B, R2, G2, B2, ..)

2. Kernel models

Evaluation

• delta E and RMSE (for spectra)– average, std, maximum

Page 36: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Spectra with largest delta E polynomial model

Page 37: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Some preliminary tests with mobile phones cameras

Page 38: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Some preliminary tests with mobile phones cameras

Page 39: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Spektrikuvan kanavakuvia

Page 40: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Spectral Face Image

Page 41: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland
Page 42: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland
Page 43: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland
Page 44: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland
Page 45: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Spectral Image

Page 46: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Image Types

TYPE SPECTRAL CHANNELS---------------------------------------• Gray-scale• Trichromatic• Spectral

– Hyperspectral• Real-time spectral

• Single• Three• >3• Numerous• Numerous

Page 47: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

MEMORY REQUIREMENTS OF IMAGES

Image size 256x256 512x512

gray-level image 65 kB 262 kB

color (RGB-) image 196 kB 786 kB

spectral, 20 nm resol. 1 MB 4 MB

spectral, 5 nm resol. 3 MB 15 MB

Page 48: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Pixels in color image are vectors

• What is the order of color?

• What means the average color?

• How to compute distance in spectral space?

• What is the structure of spectral color space?

Page 49: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Statistical Analysis of Natural Images

Page 50: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Statistical Analysis of Natural Images

Page 51: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Munsell system for color representation

Page 52: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Spectrum and Hue, Saturation, and Value

(a) (b) (c)

400 500 600 7000

20

40

60

80

100

Wavelength(nm)

Reflec

tance

(%)

400 500 600 7000

20

40

60

80

100

Wavelength(nm)

Reflec

tance

(%)

400 500 600 7000

20

40

60

80

100

Wavelength(nm)

Reflec

tance

(%)

Page 53: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

Motivation for spectral color

• Not to loose important color information

• To define optimal color sensors

• To develop better color vision models

• To develop novel instruments

• To develop spectral color classifiers and optical implementations for them

Page 54: IPCV ‘06 August 21 – September 1 Budapest Jussi Parkkinen Markku Hauta-Kasari Department of Computer Science University of Joensuu, Finland

What is color?

”Color is more than light” ”Computer cannot describe color correctly”

”Color is a perception (of human beings)”• => Color cannot been measured!• => Is this fair to animals?

• In spectral approach, color is represented by color signal. This causes the color sensation

• In spectral approach, we are not limited into this human visual range