Lecture 2 HIS color space
Last week:
introduction and tri-color representation
Digital Image Processing HSI perceptual color space
Readings: G&W 6.1-6.2, Jain 3.7-3.8 Prof. Shih-Fu Chang This week:
Review and MATLAB demo of HSI conversion Digital Video and Multimedia Lab Columbia University More Color Coordinate Systems (Jain 3.7-3.9) Jan. 27 2005 Pseudo-Coloring (G&W 6.3)
http://www.ee.columbia.edu/dvmm Optimal Quantization (G&W 2.4, Jain 4.5-4.8)
Most images are downloaded from the web site of the Gonzalez/Woods textbook Lec #2 Slide 1 Lec #2 Slide 2
Chromaticity Diagram Understanding HSI from RGB
Turn the RGB cube so that Convert 3D cube to 2D diagram Black-White axis is vertical Each plane containing the Each RGB is normalized by dividing the color values by B-W axis and any color SUM(R,G,B) point contains all the colors of the same hue (because Coordinate of each point indicates the proportion of of the mixing property) R,G,B in the composition Hue can be represented as Only shows the chromaticity of angle between the plane the color, does not specify the and a reference plane (e.g. intensity Red) r = R /(R + G + B) Boundary points are spectral Color of the same hue can colors g = G /(R + G + B) be made less saturated by Equal energy point is the white mixing more grey colors = + + color b B /(R G B) (shift closer to the b-w Points on a line are all the axis)
mixed colors using two colors Intensity can be measured on the ends by the intersection position Points in a triangle are mixed between the horizontal by colors on three vertices plane (containing the point) Æ reproducible colors by RGB and the B-W axis.
Lec #2 Slide 3 Lec #2 Slide 4
1 Formation of the HIS color cone Colors on the HIS color cone
Cross sections of the RGB Saturated colors on the cube along the B-W axis outer points The cross section shape The maximum saturation changes from triangle to values occur at the hexagon to triangle intermediate intensity Hue is represented by the levels. angle from Red line Note the different Saturation is represented equations used in the by the distance to the textbook and the reference origin book The hexagonal shape can be approximated by a (Mathematical proof on the circle or a triangle. Web site)
Lec #2 Slide 5 Lec #2 Slide 6
Manipulations in the HSI space Other Issues of Color Spaces
Uniform Color Spaces and MacAdam Ellipses (see hand notes)
Subtractive color space for printing
Pseudo Coloring Hue of Green and Blue regions are set to 0.
Saturation of Cyan is reduced HSI values of HSI allows independent by half.
Intensityprimary/secondary of the White color manipulations of colors pixels are reduced by half. colors See Matlab demo
Lec #2 Slide 7 Lec #2 Slide 8
2 Additive vs. Subtractive Colors Color Gamut
Triangle includes
Additive colors for display all the colors subtractive colors for producible by CIE printing RGB primaries See hand notes for equations Irregular region includes all the printable colors
Lec #2 Slide 9 Lec #2 Slide 10
Pseudo Coloring Reading: G&W Sec. 6.3 Intensity Slicing
How to decide the mapping function?
Lec #2 Slide 11 Lec #2 Slide 12
3 Example of intensity slicing Gray level to color transformation
Map gray level 255 to yellow
The rest to blue
255 Mapped to 8 colors corresponds Note the color variations to cracks or How to design the transformation functions? in the left lobe of the porosities in radiation test pattern the weld
Lec #2 Slide 13 Lec #2 Slide 14
Example of Gray-Color Transformation How does it work?
3 channels use sinusoidal functions with the same frequency but different phases.
Adjust the frequency to obtain the best quality/information.
Lec #2 Slide 15 Lec #2 Slide 16
4 Pseudocoloring using multiple monochrome images Sampling & Quantization
R G Reading: G&W Sec. 2.4 Jain 4.5-4.8
B R’
near-infrared
Note the difference between human- made features (blue) and the bio-mass (red)
Lec #2 Slide 17 Lec #2 Slide 18
Sampling – Spatial Resolution Quantization – Grey-Scale Resolution
See hand notes Concept of the Nyquist Rate Varying # of bits per pixel Non-uniform sampling and Interlaced sampling used in digital video
pixel replication use interpolation
Varying spatial resolution
Lec #2 Slide 19 Lec #2 Slide 20
5 Tradeoff between gray-level resolution and spatial resolution # bits
Curves show isopreference points
For low-, moderate-detail images, increasing NxN pixels K or N will help perceptual quality
For high-detail images, increasing N is more important. Lec #2 Slide 21
6