versie: 5/10/2011
Copyright notice
Didactisch materiaal bij de cursus
Beeldverwerking http://telin.UGent.be/~philips/beeldv/ Academiejaar 2011-2012
Prof. dr. ir. W. Philips
[email protected] Tel: 09/264.33.85 Fax: 09/264.42.95
UNIVERSITEIT GENT
© W. Philips, Universiteit Gent, 1998-2012
This powerpoint presentation was developed as an educational aid to the renewed course “Image processing” (Beeldverwerking), taught at the University of Gent, Belgium as of 1998. This presentation may be used, modified and copied free of charge for non-commercial purposes by individuals and non-for-profit organisations and distributed free of charge by individuals and non-for-profit organisations to individuals and non-for-profit organisations, either in electronic form on a physical storage medium such as a CD-rom, provided that the following conditions are observed: 1. If you use this presentation as a whole or in part either in original or modified form, you should include the copyright notice “© W. Philips, Universiteit Gent, 19982001” in a font size of at least 10 point on each slide; 2. You should include this slide (with the copyright conditions) once in each document (by which is meant either a computer file or a reproduction derived from such a file); 3. If you modify the presentation, you should clearly state so in the presentation; 4. You may not charge a fee for presenting or distributing the presentation, except to cover your costs pertaining to distribution. In other words, you or your organisation should not intend to make or make a profit from the activity for which you use or distribute the presentation; 5. You may not distribute the presentations electronically through a network (e.g., an HTTP or FTP server) without express permission by the author. In case the presentation is modified these requirements apply to the modified work as a whole. If identifiable sections of that work are not derived from the presentation, and can be reasonably considered independent and separate works in themselves, then these requirements do not apply to those sections when you distribute them as separate works. But when you distribute the same sections as part of a whole which is a work based on the presentation, the distribution of the whole must be on the terms of this License, whose permissions for other licensees extend to the entire whole, and thus to each and every part regardless of who wrote it. In particular note that condition 4 also applies to the modified work (i.e., you may not charge for it). “Using and distributing the presentation” means using it for any purpose, including but not limited to viewing it, presenting it to an audience in a lecture, distributing it to students or employees for self-teaching purposes, ... Use, modification, copying and distribution for commercial purposes or by commercial organisations is not covered by this licence and is not permitted without the author’s consent. A fee may be charged for such use. Disclaimer: Note that no warrantee is offered, neither for the correctness of the contents of this presentation, nor to the safety of its use. Electronic documents such as this one are inherently unsafe because they may become infected by macro viruses. The programs used to view and modify this software are also inherently unsafe and may contain bugs that might corrupt the data or the operating system on your computer. If you use this presentation, I would appreciate being notified of this by email. I would also like to be informed of any errors or omissions that you discover. Finally, if you have developed similar presentations I would be grateful if you allow me to use these in my course lectures. Prof. dr. ir. W. Philips Department of Telecommunications and Information Processing University of Gent St.-Pietersnieuwstraat 41, B9000 Gent, Belgium
E-mail:
[email protected] Fax: 32-9-264.42.95 Tel: 32-9-264.33.85
Telecommunicatie en Informatieverwerking
01b.2
versie: 5/10/2011
© W. Philips, Universiteit Gent, 1998-2012
Inleiding Beelden zijn dikwijls bedoeld om door mensen te worden bekeken
Enkele karakteristieken van het menselijk visueel systeem
Beeldverwerking wijzigt beelden, •gewild: b.v. beeldverbetering •of ongewild: b.v. compressie ⇒ Het is dus belangrijk om te weten hoe de mens afwijkingen in beelden waarneemt Het menselijk visueel systeen is uiterst complex; het bestaat uit •het oog: een sensor met lokale signaalverwerking (“preprocessing”) •de visuele cortex: lage- en hoge-niveau beeldverwerking •de rest van het brein: interpretatie Al deze deelsystemen beïnvloeden elkaar! ⇒ Hopeloos complex! 01b.4
versie: 5/10/2011
© W. Philips, Universiteit Gent, 1998-2012
versie: 5/10/2011
© W. Philips, Universiteit Gent, 1998-2012
Optische illusies
Helmholtz Stripes or hoops? Fashion myth uncovered By Steve Connor Friday, 12 September 2008
http://www.youtube.com/watch?v=WvVfcyVCdNA http://www.unoriginal.co.uk/illusionvideos.html http://www.viscog.com
01b.5
versie: 5/10/2011
© W. Philips, Universiteit Gent, 1998-2012
The first rule of fashion – that vertical stripes flatter the figure – is a myth, according to a study conducted by scientists at York University. Women who avoid wearing clothes with horizontal stripes in the belief that it makes their bum look larger should be reassured by a scientific study showing quite the opposite. It found that horizontal stripes actually make people look slimmer.
http://http://www.belfasttelegraph.co.uk/lifest yle/fashion-beauty/stripes-or-hoops-fashionmyth-uncovered-13972416.html 01b.6
versie: 5/10/2011
© W. Philips, Universiteit Gent, 1998-2012
Vermogenspectrum van licht
Golflengte en frequentie x
Opmerking: zichtbaar licht
I(x,y) dS
y
λ ≈ 550 nm (grootteorde) f ≈ 500 THz (grootteorde)
Lichtintensiteit: I(x,y)d S is het lichtvermogen dat invalt op een elementair stukje oppervlak d S (eenheid: Watt/m2) Licht kan worden ontbonden in een kleurenspectrum +∞ +∞ +∞
I ( x, y ) =
∫ I ( x, y; f ) df
=
−∞
frequentiespectrum (onafhankelijk van het transmissiemedium) 01b.7
∫
−∞
I ( x, y; c / λ )
df dλ = dλ
∫ I ' ( x, y; λ ) dλ
−∞
golflengtespectrum (gebruikt in de lichttechnologie)
In de vrije ruimte: f λ=c met c de lichtsnelheid ⇒
df c = 2 dλ λ
01b.8
versie: 5/10/2011
© W. Philips, Universiteit Gent, 1998-2012
versie: 5/10/2011
© W. Philips, Universiteit Gent, 1998-2012
Structuur van het menselijk oog
lens
Retina, kegeltjes en staafjes
Fotogevoelige elementen in de retina: •twee soorten: kegeltjes en staafjes •grootste dichtheid in de fovea •100.000.000 receptoren in het oog maar slechts 800.000 zenuwdraden ⇒ er is duidelijk lokale processing in het oog
optische zenuwbundel Staafjes en kegeltjes
Kegeltjes •aantal: 6.000.000 •voor zicht bij normaal omgevingslicht •3 types, gevoelig voor respectievelijk blauw, groen en geelgroen licht Staafjes
Zenuwbundels
•aantal: 100.000.000
⇒ voor •veel gevoeliger dan “donker”-zicht kegeltjes •1 type ⇒ geen kleuronderscheiding 01b.9
versie: 5/10/2011
© W. Philips, Universiteit Gent, 1998-2012
bipolaire cel
versie: 5/10/2011
© W. Philips, Universiteit Gent, 1998-2012
Absorptiespectra van kegeltjespigmenten
De neurale interconnecties in het oog amacriene cel
01b.10
Bipolaire cellen: •input=kegeltjes •output=ganglioncellen Horizontale cellen •maken dwarsverbindingen tussen kegeltjes en bipolaire cellen Amacriene cellen: •maken dwarsverbindingen tussen bipolaire cellen en ganglion cellen
Deze absorptiespectra geven aan hoeveel percent van het licht op een bepaalde frequentie in de kegeltjes wordt geabsorbeerd Dit licht wordt echter niet volledig omgezet in een zenuwsignaal ⇒ De absorptiespectra geven slechts een ruw idee van de spectrale gevoeligheid van de kegeltjes relatieve absorptiespectra arbitraire eenheden
retina visuele as blinde vlek fovea
β (groen) γ (geelgroen tot rood)
α (blauw) 400
horizontale cel
01b.11
(golflengtespectra!)
500
600
golflengte λ (nm)
700 01b.12
versie: 5/10/2011
© W. Philips, Universiteit Gent, 1998-2012
Spectrale gevoeligheid van de kegeltjes... Wat is het verband tussen het invallend lichtspectrum I ( f ) en de zenuwsignalen “aan de uitgang” van de kegeltjes? Experimentele vaststelling: goed model voor de respons van een kegeltje
∫
van type i op licht met spectrum I(f): f(ri ) met ri = Si ( f ) I ( f ) df met • Si( f ), i=1,2,3 de spectrale gevoeligheid van de kegeltjes van type i
versie: 5/10/2011
© W. Philips, Universiteit Gent, 1998-2012
… Spectrale gevoeligheid van de kegeltjes... Verschil spectrale gevoeligheid en absorptiekarakteristiek: •De spectrale gevoeligheidscurve Si( f ) geeft weer hoeveel van het licht met een bepaalde frequentie in een elektrisch signaal wordt omgezet •De absorptiekarakteristiek Ai( f ) geeft aan hoeveel licht bij een bepaalde frequentie wordt geabsorbeerd Verband: Si ( f ) = Ai ( f ) Ei ( f )
• f (x)≈ c log(x): een monotoon stijgende “log”-achtige functie De spectrale gevoeligheidscurve geeft weer hoeveel van het licht met een bepaalde frequentie in een elektrisch signaal wordt omgezet
Ruwe benadering: Si( f )≈ Ai( f )
De log-functie zorgt voor dynamiekcompressie (sterkte daglicht kan met een factor 20 000 variëren)
I( f )
Absorptie
Ai(f)
© W. Philips, Universiteit Gent, 1998-2012
log(⋅)
Efficiëntie Ei(f)
versie: 5/10/2011
f(ri )
Dynamiekcompressie 01b.14
© W. Philips, Universiteit Gent, 1998-2012
...Spectrale gevoeligheid van de kegeltjes
Metamerisme relatieve absorptiespectra
Ruwe benadering voor Si’(λ ): Si’(λ)≈ Ai’(λ )
β (groen)
β (groen)
γ (geelgroen
γ (geelgroen tot rood)
tot rood)
α (blauw)
α (blauw) 400
fotonen → elektronen
Spectraal filter/fotocel
01b.13
versie: 5/10/2011
ri = ∫ Si ( f ) I ( f ) df
bron2 bron1
500 600 golflengte (nm)
700
Conclusies: • Het oog is weinig gevoelig voor blauw • De gevoeligheidscurven overlappen vrij veel ⇒ zelfs bronnen met een smal spectrum zullen meestal meerdere types kegeltjes activeren Opmerking: Si’(λ) =Si (c/λ) 01b.15
400
500 600 700 golflengte (nm) Metamerisme: bronnen met een verschillend spectrum kunnen de zelfde kleurindruk geven (de mens ziet geen verschil tussen de bronnen) Logisch: I( f ) is een functie en dus bepaald door oneindig veel getallen en er zijn maar drie soorten kegeltjes ⇒ er gaat informatie verloren Er bestaan minder triviale voorbeelden van metamerisme (zie appendix!)
01b.16
versie: 5/10/2011
© W. Philips, Universiteit Gent, 1998-2012
versie: 5/10/2011
© W. Philips, Universiteit Gent, 1998-2012
De wet van Weber... Kleinst waarneembaar contrast: (grijs licht)
I I + ∆I
∆I / I ≈ 0.02
∆I I
(Weber’s wet) 2%
Intensiteit I
Deze experimenteel vastgestelde wet suggereert een logaritmisch verband tussen de “waargenomen helderheid” I’ en de opgenomen lichtintensiteit I:
∆I ' = 1 ↔ ∆I ≈ 0.02 I 1 ∆I ' ⇒ = ⇒ I ' = 50 ln( I ) + b ∆I 0.02 I ⇒ I ' = a log( I ) + b Alhoewel een logaritmisch verband het experiment grotendeels kan verklaren blijkt dat de werkelijkheid veel complexer is en dat de “waargenomen helderheid” I’ in een punt ook afhangt van de lichtintensiteit in andere kegeltjes 01b.17
...De wet van Weber Kleinst waarneembaar contrast tussen 2 lichtvlekken op een achtergrond Besluit:
I0 I + ∆I
I
∆I / I ≈ 0.02 f ( I − I 0 ) I0 = C1
I0 = C2
∆I I
•Weber’s wet gaat maar op als I≈I0 •Intensiteitsverschillen tussen twee gelijkende lichtintensiteiten I en I+∆I zijn moeilijker waar te nemen als de intensiteit I0 van achtergrondlicht sterk verschilt van I •de waargenomen helderheid van een punt hangt ook af van de lichtintensiteit in andere punten Verklaring: •het oog past zich aan het achtergrond-licht (omgevingslicht) aan •cfr. verblinding door de zon
Intensiteit I
01b.18