Aberrations Caused by Decentration in Customized Laser Refractive Surgery

1
An Opponent Process Approachto Modeling the Blue
Shift of the Human Color Vision System
Brian A. Barsky, Todd J. Kosloff, and Steven D.
Upstill1Computer Science Division, University of
California, Berkeley, California, 94720-1776,
U.S.A. 1Box Rocket Animation Ltd. Wellington,
New Zealand
THE OPPONENT PROCESS THEORYOF COLOR VISION
RANGE OF POSSIBLE BLUES
BACKGROUND
ALGORITHM

• Our model requires that the blue shift be along
  the x y chromaticity line.
• Hunts blue point, (.3, .3), happens to be on
  this line, providing support for our model.

• Dimly lit scenes appear desaturated and bluish
• Low light levels have a variety of effects on
  human visual perception. Specifically, visual
  acuity is reduced and scenes appear bluer,
  darker, less saturated, and with reduced
  contrast. Presently, we confine our attention to
  an approach to modeling the appearance of the
  bluish cast in dim light, which is known as blue
  shift. There is physiological and psychophysical
  evidence to explain this blue shift. Both
  photographs and computer-generated images of
  night scenes can be made to appear more realistic
  by understanding these phenomena and taking them
  into account.

• Convert input image into CIE XYZ color space
• Convert from XYZ space into opponent color space
• Desaturate image by attenuating chromatic
  channels
• Add blue tint by shifting the yellow/blue
  channel
• Lower brightness by attenuating achromatic
  channel
• Convert output into XYZ and then RGB space

• Color is encoded as three opponent channels
• yellow/blue, red/green, and white/black
• The trichromatic theory of color vision explains
  that all colors are combinations of responses to
  the three cones.  However, the physiologist Ewald
  Hering showed that the trichromatic theory does
  not adequately explain various phenomena. He
  noted the specific correspondence of the color
  that appears in an after-image after extended
  viewing of a particular color (red after green,
  blue after yellow, and vice versa).  He also
  observed that there are certain pairs of colors
  that we never see as occurring together
  (red/green, yellow/blue,) that is, that no
  colors appear to be "reddish green" nor "bluish
  yellow", even though there are colors that appear
  as "yellowish green", "bluish red", or "yellowish
  red".   The theory was validated and quantified
  in the 1950's at Eastman Kodak by Leo Hurvich and
  Dorothea Jameson 1957. 
• Unlike the trichromatic theory, which operates
  at the receptor level, the opponent process
  theory applies to the subsequent neural level of
  color vision processing. The signals are
  neurally interconnected in three channels, each
  comprising an opponent pair (that is,
  blue/yellow, red/green, and luminance).  For each
  pair, activation of one member inhibits the
  opposing member. 

Y
CIE Chromaticity Diagram

• Vision science explains these effects
• The retina comprises two kinds of photoreceptors
  called rods and cones. The rods are more
  sensitive in dim light than are the cones. In
  very low light levels where the rods are
  responsive but the cones are not, vision is said
  to be scotopic. Conversely, photopic vision
  occurs when the level of illumination is too
  bright for rods but is suitable for cones.
  Furthermore, mesopic vision refers to the
  intermediate stage where both rods and cones are
  active however, the cones are not as sensitive
  as they are at higher light levels, so the scene
  appears desaturated. Although there are three
  different kinds of cones with different spectral
  sensitivity curves, all rods have the same
  spectral response curve. Consequently, rods
  provide luminance information but no color
  discrimination. Thus, when the light is too dim
  to excite the cones, scenes appear monochromatic.
  Trezona 1970 theorizes that blue-shift is a
  result of rod-cone interaction.

where knight is a number between zero and one
controlling the amount of desaturation, and
kblue determines how much blue is added to the
image. V represents the scotopic luminosity
that is rod response. CWB corresponds to
photopic luminance. Unfortunately, most
computer generated images do not provide both
photopic and scotopic luminance values for each
pixel. We approximate rod response via the
linear transformation given in Pattanaik 1998.
(1/3, 1/3)

(.3, .3)
XY
PREVIOUS WORK
L
M
S
PSYCHOPHYSICAL CALIBRATION
Short, Medium, and Long Wavelength Cones

X

• A user may desire a different shade of blue
• Any realizable x y below the CIE white point
  (1/3,1/3) may be chosen.

• We build on our 1985 blue shift model
• That model depended on user-defined values
• In 1985, we produced a comprehensive
  computational model incorporating the sensitivity
  of receptors to luminance differences, spatial
  information processing, in which the output of a
  single neuron is influenced in important ways by
  many receptors across the retina, and the
  mechanism for encoding color, and the nature of
  the color representation. Upstill 1985
• One portion of our model processed images to
  appear as if they were viewed at night by
  desaturating and blue-shifting the image. Our
  model was based on theories from the vision
  science community on the nature of the retinal
  mechanisms, but we avoided precise calibration of
  the various mechanisms, instead allowing the user
  to calibrate the model so as to produce the most
  pleasing images. Presently, we calibrate the
  color processing portion of our model. We bring
  in data from Hunt 1952, and show how this data
  can be incorporated into our model, producing
  realistic results.
• Our formulation derives from vision science
• Over the past few years, we have seen a few
  different models of blue shift appear. Most
  recently, Thompson et al. presented a technique
  which, like ours, produces night images by taking
  into account desaturation and blue shift
  Thompson 2002. Also like us, they used Hunts
  psychophysical data to calibrate their blue
  shift. However, Thompson transformed colors
  using an ad-hoc set of equations, whereas our
  formulation derives from physiological knowledge
  of the human visual system Hurvich and Jameson
  1957.

• The parameters knight and kblue can be provided
  to the user to set as desired for best visual
  effect.
• However, these parameters can also be set by
  experiment.
• In Hunt 1952, several observers participated in
  an experiment to precisely
• determine the appearance of colors to a dark
  adapted eye. Hunt observed that
• as the adapting light was decreased to zero, all
  perceived color differences were
• lost, and all hues converged to a single percept
  with chromaticity coordinates
• equal to approximately x0.3, y0.3. We
  incorporate this experimental result
• into our blue-shift model.
• We set knight and kblue so that in the extreme
  case of complete dark adaptation,
• the image is monochromatic, with a chromaticity
  of (0.3, 0.3). Setting knight to 1
• makes the image monochromatic. We find
  chromaticity by looking at our model
• in terms of XYZ coordinates.

B Y -
G R -
Wh Bk -
Neural Opponent Response
PARTIAL BLUE SHIFT
ALGORITHM SCHEMATIC

• Fully monochromatic, blue-scale images
• are not necessarily desirable
• Intermediates between the original and fully
  
• blue shifted images are easily created
• knight can be manually adjusted to set the amount
  by which an image
• should be shifted. We automatically scale kblue
  by knight to ensure that the amount of blue is
  proportional to the amount of desaturation and
  dimming. Thus we achieve a continuous blend
  between the original image, and an image in
  accord with Hunts experiment on complete dark
  adaptation.

kblue
Scotopic Luminance

knight
1 knight

Output
Input

X
Bk/Wh
Opponent Response
Bk/Wh

Y
B/Y

B/Y
R/G
Z
R/G