Transcript Digitization of Images

Digitization
Informatics INFO I101
January 26, 2004
John C.Paolillo
Mars “Anomalies”
• Are the photos from various Mars missions
faked?
• Do the existing photos show evidence of
civilized life?
• Do the photos show evidence of martians?
Etc.
http://rush.digitalchainsaw.com/marspath.html
Raster Image Digitizaztion
• Grid
– Resolution
• Quantization
– Bit-depth
– Color
cf. Vector Graphics
1280  960
Great Sand Dunes, Colorado (IRS satellite image; www.spaceimaging.com)
640  480
320  240
160  120
80  60
256 Grays
16 Grays
4 Grays
Black and White
Digital Artifacts
• Pixelation (“jaggies”)
– From discretization of the analog signal
• shape
• color/gray level
• Resolution mismatches
– cause geometric distortions as error accumulates
• Fix: digital interpolation
– dithering
– anti-aliasing (especially with fonts)
Color
Color Perception
3 types of retinal
sensor cells, sensitive
to 3 different bands of
light
3 Electron guns, aimed
at 3 different colors of
phosphor dots —
analog signals
Color: Response Patterns
green cones
red cones
Wavelength 
blue cones
The Eight-Color World
• Eight colors: black, yellow, magenta, red,
cyan, green, blue, white
• Three color tubes on a TV monitor: Red,
Green, Blue
23=8
• Additive color relations:
red+green+blue=white
A Psycho-Physical Encoding
RGB
101
100
110
000
Wavelength 
010
011
001
111
101
More Colors
• Recognize more levels in each channel
– 2 bits per channel: 26 = 64 colors
– 4 bits per channel: 212 = 4096 colors
– 8 bits per channel: 224 = 16,777,216 colors
• Except for 3-bit and 24 bit colors, most standard
colors are not in multiples of 3 bits
– 8 bits (256 colors)
– 16 bits (32,768 colors)
(8 bits is a convenient storage unit)
The Color Table
• A table of some convenient number of values
– 4, 16, 256, etc.
• Each location in the table is mapped to some
higher resolution color value (24 bit)
– Some locations may be unused (mapped to black)
• A monitor typically uses only one color table at a
time
Signal Levels
• Intensity is analog
• Levels are digital
How do we convert analog intensity to
digital levels?
• Quantization: convert analog signals to
digital numbers
Quantization
1
1
0
1
0
0
1
White
111
0
110
1
101
0
Med
Gray
100
1
011
0
010
1
001
0
Black
000
1. Evenly divide signal levels
2. Assign a unique binary
number to each recognized
level
3. Match signal with recognized
levels and round any
intermediate signal level to
the nearest recognized level
4. Report the signal as a list of
binary numbers
Counting in Binary
• Two values: 0, 1
• Each digit is a power of 2
– 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024, ...
– Fractions: 0.5, 0.25, 0.125, 0.0625, 0.03125, ...
– positive, negative, rational, real, imaginary...
We’ll stick to whole numbers for now
Binary counting
• Start with zero:
00000
• Add 1:
00001
• Adding 1 more carries:
00010
• Add 1:
00011
• Adding 1 more carries 2x:
00100
• Add 1:
00101 Etc.
OR:
• Divide the full range into 2 halves, 0 (low) and 1
• Divide each range again for each next bit
• Stop with the last bit
End