Simple model: simple reflectance/illumination model Eye illumination source i(n 1,n 2 ) image: x(n 1,n 2 )=i(n 1,n 2 )r(n 1,n 2 ) reflectance term r(n 1,n 2 ) where 0 < i(n 1,n 2 ) < 0 < r(n 1,n 2 ) < 1
Imaging on the retina (back of eye consisting of photoreceptors) Focal point of lens Eye 2m Retinal image 17mm 100m 20mm
Visible range of electromagnetic spectrum is 350 nm to 780 nm. 380nm 780nm rays x rays ultraviolet visible infrared microwaves radio
Simple model for HVS eye brain optic nerve HVS Input (spatial pulse) What we see Approximate HVS with a LTI system HVS Primarily a BPF/LPF Output What we think we see NOTE: The HVS is really a non-linear system.
Light: electromagnetic radiation that stimulates our visual response expressed as a spectral energy distribution C( ); 380nm 780nm wavelength in visible spectrum Spectral distribution of a colored light C( ) represents amount of energy present at each frequency wavelength
Color vision model (3 receptor absorption model) 3 types of cones: each has a different peak absorption frequency Typical absorption spectra (also called sensitivity curves) for the three cones (not to scale)
Let C(): spectral energy distribution of a colored light source C() min max max
Color sensation described by i max C S C d, i 1,2, 3 min i i [C()], i=1,2,3, called spectral responses If C 1 () and C 2 () produce responses such that i [C 1 ()]= i [C 2 ()] for i = 1,2,3 C 1 () and C 2 () perceived to be identical Color sensation perceptual attributes 1. Brightness perceived Luminance 2. Hue color 3. Saturation amount of white light diluting the color
Fraction of light absorbed by each type of cone Vision & Perception The following curves show the relative spectral response functions of each of the 3 types of cones: 0.2 G R 0.01 0 B 400 550 700 (nm) Eye s response to Blue light is much less strong than is its response to Red or Green
Luminance and Brightness The luminance or intensity of an object with light spectral distribution I(x,y, ) is L 0 x y I( x, y, ) V, d where V( ) is the relative luminous efficiency function of the HVS Bell-shaped curve = sum of 3 previous curves.
Luminance versus Brightness Brightness: subjective perceptual measure (depends on observer s judgment) perceived Luminance depends on luminance of the surround (lateral inhibition, contrast) Luminance: objective quantitative measure (Unit: watts/m 2 or watts/steradians) independent of the luminances of the surrounding objects Note: the illumination (Luminance) range over which HVS can operate is roughly 10 10 (normalized unit, e.g. milli-luminance = milli-unit of luminance) or 10 orders of magnitude on log scale.
Scotopic vision mediated thru rods at the lower part of the range Photopic vision mediated thru cones at the higher 5 to 6 order of magnitude of interest here, computer screens are bright Our perception is sensitive to luminance contrast rather then the absolute luminance value. Brightness is log related to luminance Brightness (log scale) Glare limit scotopic photopic Brightness is approximately linear on the log scale Scotopic threshold 10-6 10-1 10 3 milli-luminance (power)
Concept of just-noticeable difference (contrast sensitivity) Experiment: Human observer views background L and a spot with intensity L+L. As we change L, dot becomes visible. The L for which dot is visible is the just-noticeable difference. L L+L L/L is the Weber ratio Weber s law: L/L= C (constant) = 0.02 d(logl) = constant C equal increments in log L should be perceived to be equally different ( linear relation between Brightness and log L) log L is proportional to C, the change in contrast
10-1 10 3 milli-luminance
Exploit this brightness property to derive contrast models: c = a 1 + a 2 log f logarithmic law c = f 1/n root law
Visual Acuity Ability to detect spatial details; spatial frequency sensitivity of the eye
Retinal arc Divide the eye into degrees 20mm 17mm 30 o 30 o 10 o 0 o 10 o Images are projected onto rods and cones by retinal arc. We can unwrap the retina:
Spatial frequency is not related to the wavelength of the light is the number of oscillations in a given space 30 o 10 o 30 o 0 o 10 o 0 o 1 o 4 cycles/retinal arc
Color Models RGB CIE spectral primary sources; CRT monitors CMY Printers; ink-based devices Traditionally, RGB primary colors, CMY complements of RGB C = W - R M = W - G = R + B Y = W B = R + G R N G N B N NTSC receiver primaries; standard for television receivers; three phosphor primaries that glow in the red, green, and blue regions of the visible spectrum YIQ NTSC transmission standard; compatible with B/W TV broadcast; more efficient transmission than RGB HSV or HSB User-oriented, based on intuitive or perceptual measure Note: NTSC stands for National Television Systems Committee
Color Models RGB (CIE primaries) color matching functions T B ( ) T R ( ) T G ( ) The tristimulus values (weights) of an arbitrary color C( ): t k max min C( ) T k d k R, G, B
Color Models CIE Chromaticity Diagram CIE defined 3 standard (hypothetical) primary sources called X, Y and Z to replace R,G and B. These new primaries can match all visible color with positive weights (positive matching functions) Y color matching function matches the luminous efficiency function of the eye 2 1.8 z n 1.6 1.4 1.2 1 y n x n 0.8 0.6 0.4 0.2 0 380 410 440 470 500 530 560 590 620 650 680 710 740 770 (nm)
Color Models Let Then C C xx ' axx yy zz ayy azz will produce the same color but with a different intensity; i.e., same Hue and Saturation, but different Brightness Normalize by setting C n x n X a x y n Y y z z n Z where x n x y ; yn x y z x y z ; z n x z y z
Color Models Note: z n xn yn z 1 x y n n n 1 (Unit Plane) out of the 3 normalized weights, only 2 have to be specified only 2 primaries needed to define color CIE diagram = projection of Unit Plane into (X,Y) plane x n y n The three values, and define hue and saturation but give no info about the brightness since they are relative components An extra value is required to determine the intensity (Brightness) and the value of Y is chosen, In practice, any absolute intensity value (x, y or z) may be specified to determine the brightness. z n
Color Models CIE Chromaticity Diagram y n Curve (Horse-shoe) boundary corresponds to 100% pure colors 546.1nm G All possible colors (of normalized intensity) are displayed on CIE diagram 0.333 The (MacAdam) ellipses are the just noticeable color difference ellipses. White: x n = y n = 0.333 B 435.8nm White 0.333 Yellow 700nm x n R z n = 1 x n y n = 0.333
Color Models YIQ: NTSC transmission standard Y = Luminance (same as CIE Y primary); color matching function identical to luminous efficiency function V( ) I and Q: chrominance components (give hue and saturation) Recoding of R N G N B N for transmission efficiency Transmission efficiency: Bandwidth of I or Q < half bandwidth of Y NTSC encoding of YIQ into a broadcast signal assigns: 4 MHz to Y 1.5 MHz to I 0.6 MHz to Q I and Q components contain less information less samples (more than 50%less) used to represent I and Q Downward compatibility with B/W TV receivers (Y component)
Color Models Converting R N G N B N to YIQ: Y 0.299 I 0.596 Q 0.211 Recall: L 0.587 0.274 0.523 max min C( ) V 0.114 R 0.322 G 0.312 B d C( ) consists of only three components of weights R N at R, G N at G and B N at B C( ) = R N ( - R ) + G N ( - G ) + B N ( - G ) becomes a summation weighted by the corresponding V( R ), V( G ) and V( B ) Y = L() = V( R ) C( R ) + V( G ) C( G ) + V( B ) C( B ) Colored light distribution = 0.30 R N + 0.59 G N + 0.11 B N N N N Y 0.30R 0.59G 0. 11B C() B N B G N R N G R
Color Models Some useful transformations between color coordinate systems RGB to XYZ X Y Z 0.490 0.177 0.000 0.310 0.813 0.010 0.200R 0.011 G 0.990 B R N G N B N to XYZ R G B N N N 1.910 0.985 0.058 0.533 2.000 0.118 0.288 X 0.028 Y 0.896 Z
Temporal Properties of Vision Important for processing motion images (video) and in the design of image displays for stationary images Main properties: Bloch s law If we expose an observer to flashing light where flashes have different durations but same energy these durations became indistinguishable below a critical duration threshold d 1 d 2 Flash 1 duration d 1 indistinguishable of d 2 Flash 2 duration if d 1 d c and d 2 d c This threshold was found to be about 30 ms when eye adapted at moderate illumination level The more the eye is adapted to dark, the longer the critical duration
Temporal Properties of Vision Critical Fusion Frequency (CFF) If flashing rate of light > CFF individual flashes are indistinguishable; i.e., flashes are indistinguishable from a steady light at the same average intensity CFF does not generally exceed 50 to 60 Hz Basis for TV raster scanning cameras and displays Interlaced image fields sampled and displayed at rates of 50 or 60 Hz Modern displays are refreshed at 60 frames/sec to avoid flicker perception
Temporal Properties of Vision Spatial versus Temporal effects: Eye more sensitive to flickering of high spatial frequencies (i.e. flickering edges) than low spatial frequencies Useful in coding of motion video where moving areas are subsampled except at the edges (low spatial areas represented by less samples)