Color Appearance Models

Transcript

Color Appearance Models
Color Appearance Models Sensazione e percezione •  Colorimetria: misura. •  which paint should a painter use to depict an image of the ra;? •  Sensa=on •  which paint should be used in order to re-­‐
paint the ra;? •  Percep=on Apparenza •  Sensa=on is connected with appearance. •  Same color s=mulus in different context: different appearance •  Simula=ng the appearance is the objec=ve of some computa=onal models called colour appearance models (CAMs). Illusioni oHche • Color/luminance illusions -­‐ Something that deceives our HVS -­‐ Can reveal mechanisms of percep=on • Examples -­‐ Simultaneous contrast -­‐ Color constancy Contrasto simultaneo Adelson’s illusion Costanza croma=ca • Ability of the HVS to perceive the color of the objects almost the same (even changing illumina=on condi=ons) Colorimetria CIE •  Basic colorimetry provides a measurement of the colours observed in isolated viewing condi=on. •  (I due cuscini in ‘b’ e ‘c’ sono uguali perchè hanno gli stessi XYZ o gli stessi RGB) •  In everyday life, colours are rarely seen in isola=on; rather, they are seen in complex scene •  Therefore, it originates the needs of an advanced colorimetry (I due cuscini non ci appaiono uguali) CAMs •  Dis=nguiamo diverse famiglie che chiamo con un termine generale Color Appearance models •  Famiglia derivante da studi connessi in qualche modo alla CIE •  Famiglia derivante dalla teoria Re=nex Color Appearance Models •  (Hunt, 1982), (Nayatani, et al., 1987) and the CIE models CIECAM97s (CIE, 1998), CIECAM02 (Moroney, et al., 2002) •  Alcune idee riguardan= tali modelli Viewing condi=ons •  The colour appearance of a s=mulus is strongly related with the surrounding area •  To use a colour appearance model it is necessary to define the viewing field, that is the environment in which a s=mulus is viewed Viewing field Everything outside the s=mulus, therefore the set of proximal field, background and surround, is called the adap=ng field. Chroma=c Adapta=on •  Colour constancy: the human visual system is able to perceive the colours of the objects even under a change in the illumina=on condi=on. •  One explana=on to this behaviour is known as chroma=c adapta/on, that is the capability to adjust the general sensi=vity to preserve the color appearance of the objects. •  Esiste un adahamento alla luce/buio: fisiologico •  Adahamento o non adahamento? Questo è il problema… Corresponding color •  Ho bisogno di da= •  Through a mul=plicity of experiments, a variety of corresponding-­‐colour data have been obtained (Hunt, 1952), (Helson, et al., 1952), (MacAdam, 1961), (Wright, 1981), (Mori, et al., 1991), (Fairchild, 1991), (Luo, et al., 1991a), (Luo, et al., 1991b), (Hunt, et al., 1994). Corresponding color •  when two different s=muli, viewed under different viewing condi=ons, match in appearance, represent a pair of corresponding colours. •  cerchi: D65 (luce naturale) •  Triangoli: A (tungst) Metodi sperimentali •  Haloscopic: refers to those experiments in which one eye is adapted to one viewing condi=on and the other one is adapted to a different viewing condi=on. Then two s=muli are presented separately to the eyes in order to verify if there is matching (Mori, et al., 1991). •  Memory matching: the observer looks to a s=mulus in a viewing condi=on, then adjust another s=mulus with different viewing condi=on un=l they match (Helson, et al., 1952), (Wright, 1981), (Fairchild, 1991). •  Magnitude es=ma=on is an experiment that assigns a value to various type of appearance ahributes as lightness, chroma, hue etc (Luo, et al., 1991a), (Luo, et al., 1991b), (Hunt, et al., 1994). CAT •  The first step towards the building of a CAM is the chroma=c adapta=on transform (CAT). •  A CAT is able to transform the tris=mulus values observed under certain viewing condi=ons to matching tris=mulus values observed in a second set of viewing condi=ons. •  Therefore a CAT can be used to make predic=ons of corresponding colours Ma come CAT…? •  Transform the CIE tris=mulus values (XYZ1) into a cone response domain (LMS1). •  Predict the adapted cone signals (LMSa) based on the ini=al viewing condi=ons (VC1). •  Determinate the cone excita=on (LMS2) according to the second set of viewing condi=ons (VC2). •  Back to tris=muls values (XYZ2). •  Ciò significa che è necessario conoscere sia le condizioni di partenza che quelle di arrivo Workflow •  Mol= modelli di apparenza, stessa idea: •  Input data: tris=mulus values (XYZ) for the s=mulus, data concerning the viewing environment, background, surround, light source. •  A transforma=on from XYZ to cone response, to beher model the physiological processes of the HVS. •  A chroma=c adapta=on transform is performed. •  Some opera=ons in order to model the opponent theory of colour vision. •  Output: appearance ahributes including lightness, hue and chroma at least. •  Vedi ciecam02 da pag 15 – mostrare img F_La Fahore D •  Prima di fare adahamento croma=co si può scegliere il fahore di adahamento D. •  Se vale 1: adahamento completo •  Se vale 0: non c’è adahamento (in pra=ca non si può avere 0) •  Esempio: plug-­‐in Photoshop •  NOTA: devo dargli l’illuminante in input, non predice effeH locali iCAM •  A colour appearance model ahempts to es=mate the appearance of a single s=mulus observed on a certain uniform background, usually specified simply by its luminance (no chroma=c). •  Not useful to study the appearance of an image. iCAM •  A par=re dal 2002 Fairchild comincia a sviluppare un nuovo modello che lavori su immagini. •  iCAM nasce come tone mapper per immagini HDR, estendendo il modello CIECAM02 •  Il modello vuole quindi in input una immagine HDR e nient’altro (o quasi). Workflow •  The image is separated in two images: base layer and detail layer (bilateral filter ). •  The next steps can be applied only on the base layer, to preserve the details. •  A blurred version of the image is calculated: white layer •  Compressioni alte luci e toni scuri (curva a S) CAT •  Per fare l’adahamento croma=co si divide ogni punto per l’immagine per il proprio punto di bianco (white layer) •  D=1: R=R/Rw •  D=0: R=R •  Località: punto di bianco diverso per ogni pixel Esempio (230 cost.) Esempio Re=nex based •  Modelli: RSR, STRESS, ACE •  Caraheris=ca interessante (differenza coi CAM): non è necessario conoscere a priori l’illuminante •  Comportamento locale e non globale •  Lavorano su immagini