Digital Camera

Essential ElementsPart 2

Digital Image Quality

• Image Quality - more comprehensive than resolution and more meaningful

• 4 Pillars of Quality– http://www.normankoren

.com/Tutorials/MTF7.HTML#PillarsResolution- a function of: lens+pixel number+file

processing softwareNoise - analogous to grain in film photographyDynamic Range – Maximum signal / Minimum “usable”

signal- usually expressed in powers of 2 (f-stops)Color Accuracy – White Balance, Saturation,

Digital Camera Resolution

• Megapixels – only one of many factors that limit resolution

• Other factors– Pixel size or pitch– Lens quality– Interpolation software– System noise (electronic and photon)– Image data file compression

Resolution

• http://www.dpreview.com/learn/?/Glossary/Digital_Imaging/Resolution_01.htm

• Measured in line pairs / mm– Vertical (primary), Horizontal & – High contrast bar chart input– Low ISO (low noise) exposure– In camera processed, then printed

Digital Noise• http://clarkvision.com/imagedetail/does.pixel.s

ize.matter/• CCD Noise – Photon “Shot” noise

– Photons arrive at random intervals, so there is some uncertainty about the “true number” collected

– Photon Noise: (# Photons collected)1/2

• Total Noise: ((Photon Noise)2+(Electronic Noise)2)1/2

• Effective S/N: #Photons/Total Noise– Photon noise is dominant in a usable signal

(2) Digital Noise Cont’d

• Why do I care about noise?– Because we see it as luminance variation

in uniform areas within the photo– ISO 12232.1998 standard

• S/N >~10: “acceptable” quality for 18% gray patch (R,B,G = 46)

– S/N >~32: “noise” not visible– S?N >0.3: white patch just visible on black

background

Digital Noise ExampleS/N = 10 @ 18% gray (R,B,G=46) (~219 photons)

Digital Noise ExampleS/N = 4 @ 18% gray (R,B,G=46) (~73 photons)

(3)Digital Noise Cont’d

• Noise standard sets a ”floor” on the number of photons that need to be collected– Signal:219, Shot Noise:14.8, Read noise:16, – Total noise: (475)1/2 = 21.8

• S/N = 219/21.8 =10• CCD is a linear amplifier, so (256/46)x219=1219 photons

is minimum “full well” at level 256 for an “acceptable” image, at least from the stand point of noise.

(4)Digital Noise Cont’d

• Oops!– The inherent ISO of the CCD is ~50, but

most cameras set the default ISO to 100• Therefore we need to double the full well

capacity of the pixel to 2438 photons in order to maintain the minimum acceptable noise standard

(5)Digital Noise Cont’d

• High ISO – aka photon multiplier– Since fewer photons are counted to

create a given luminance value, the S/N is also lower compared to a lower ISO setting

Digital Noise Conclusions

• Larger pixels enable more photons to be collected > higher S/N

• Shadows > lower S/N

• High ISO > lower S/N

Dynamic Range

• Definition: Maximum photon capacity / minimum (usable) photon count.

• Maximum photon capacity is limited by pixel size – small pixels collect fewer photons

• Minimum (usable) photon count is noise limited

(2)Dynamic Range Cont’d

• It is convenient to discuss dynamic range in terms of f-stops, i.e doubling the exposure >1 f-stop, or “zone” .

• Printed images are limited to ~ 5 1/2 zones• A good monitor is limited to ~ 8 zones • Real scenes may exceed 10 zones• 8 bit/ color digital file can contain 8 zone

exposure range (if the camera can deliver!)

(3)Dynamic Range Cont’d

• Pixel photon collection requirements– 6.02n +1.8 @ pixel saturation

n= digital bits /channel8 bits required for full range jpg image, but a 10 bit A-D converter is used to “fill” the the shadows when the tonal correction is applied to linear data collected.

Therefore n=10 in the equation above

S/Nmin = 62 or 4100 photons @ saturation – vs. 2438@ saturation for acceptable noise

Minimum Pixel Size

• Current CCD (or CMOS) collect ~835 photons/ micrometer2

– Minimum pixel area = 4.9 micrometer2

– For square pixels: pitch = (4.9)1/2 =

2.2 micrometer

• Cameras with pixels this small have barely enough dynamic range for a JPG photo and will require noise reduction for any ISO greater than 100

(4)Dynamic Range cont’d

• JPG is limited to 8 bits/channel but the actual dynamic range is less, depending on the A-D converter and the “noise floor”

• After A-D conversion, the signal from the linear CCD undergoes “tonal correction” to match our eye response : Sout = K x (Sin)1/Gamma

Gamma ~ 2

K = 16 for 8bit Sout

(5)Dynamic Range cont’d

• A-D converter & Tone conversion have a big effect on dynamic range– 10 bit A-D after tone conversion allows only 5 stops,

because the first non-zero level is 8.• 256/8=32 = 25

– 12 bit A-D allows 6 stops– 14 bit A-D allows 7 stops

Tone Correction

Tone Correction from 8 Bit A-D Converter

Tone Correction from 10 Bit A-D Converter

Noise & Dynamic Range (before tone conversion)

3.4 micrometer pixel pitch

Zone # Photons Photon Noise Read Noise Total Noise S/N db8 10000 100 16 101 99 207 5000 71 16 72 69 186 2500 50 16 52 48 175 1250 35 16 39 32 154 625 25 16 30 21 133 312 18 16 24 13 112 156 12 16 20 8 91 78 9 16 18 4 6

(entry level camera)

CCD with 10,000 photon capacity

Noise & Dynamic Range (before tone conversion)

5.8 micrometer pixel pitch

Zone # Photons Photon Noise Read Noise Total Noise S/N db10 28000 167 16 28256 168 167 229 14000 118 16 14256 119 117 218 7000 84 16 7256 85 82 197 3500 59 16 3756 61 57 186 1750 42 16 2006 45 39 165 875 30 16 1131 34 26 144 438 21 16 694 26 17 123 219 15 16 475 22 10 102 109 10 16 365 19 6 81 55 7 16 311 18 3 5

CCD with 28,000 photon capacity (consumer SLR camera)

Color Quality

• How accurately does the camera interpret the color in the image?– RBG filter cut-off – Bayer Interpolation accuracy– White Balance– “Over flow” artifacts – Tone Curve

Color Quality Cont’d

• RBG filter cut-off– Bayer filters well controlled – Foveon filter - technical challenge since the

filter/sensors are “stacked”

• Bayer interpolation – How many surrounding pixels are “read” to determine RBG channel luminance?– Circle of confusion &/or diffraction means the

image information spans several pixels. – Interpolation is computation intensive > camera

computer is limited vs. PC

Color Quality Cont’d

• http://www.dpreview.com/learn/?/Glossary/Digital_Imaging/White_Balance_01.htm

• White Balance – How well does the camera “gauge” the luminance distribution– Camera software is remarkably ingenious, but

not perfect and there may be significant differences between cameras manufacturers and across the spectrum of luminance distribution.

Color Quality Cont’d

• Overflow Artifacts– When a pixel is over-exposed, excess

electrons may “overflow” to adjacent pixels causing color fringing, especially in high contrast areas

– Inherent to CCD construction

Color Quality Cont’d

• Chromatic Aberrations– A function of lens design: red, blue and

green light may not focus at same point in the image.

– After interpolation, the effect is very similar to color fringing and most obvious in high contrast areas farthest from the lens axis

Color Quality Cont’d

• http://www.normankoren.com/digital_tonality.html#Raw_conversion

• CCD is a linear device, but we “see” as a log function– CCD: luminance proportional to exposure

– Visual: Log10 luminance proportional to

Log10 exposure

• Tone Curve “converts” from CCD to visual

Power Source

• Batteries – Alkaline - not rechargeable, readily

available– Ni-MH – Rechargeable & available to

replace alkaline– NI-MH – custom form factor unique to

camera– Lithium Ion – custom form factor unique

to camera

Illuminator

• Electronic Flash is universal– Fixed location (entry level)– Elevated operating position –improved

red eye– Hot shoe – to accept more powerful,

more flexible flash ( articulated, power modulated)

Focus

• Auto focus is standard on all but the most basic (fixed focus) cameras

• Most cameras have auxiliary lamp to aid focus operation in low light ambient

• More expensive cameras offer multiple focus programs (macro, portrait, landscape) + manual focus.

Exposure Control

• Digital Camera exposure latitude is far less than print film (and about equal to slide film). Auto exposure adjust is universal– More expensive cameras offer aperture

dominant + shutter speed dominant exposure modes

– More expensive cameras offer multiple “scene averaging” programs

Aiming Devices

• LCD (fixed or articulated) • Auxiliary optics (point & shoot)

Parallax error ~80% true field of view• Electronic view finder (LCD + lens)

– Accurate field of view >98%– Relatively low resolution, both color and detail– Supplementary data (histograms, shooting

data)

Aiming Devices Cont’d

• Reflex (SLR)– High end cameras with multiple lens

options– Accurate field of view– Excellent color and detail– LCD view mode not available until “after

the shot”

Data Storage

• Fixed in-camera – Entry level

• Removable flash memory– Type 1, Type 2 compact flash (CF) cards– SD, XD & Memory stick (all smaller than

CF cards)

Summary

• Camera Properties– Ergonomics- Does it fit your hand/eye?– Controls- Intuitive vs. Learning Required

• Adjustments for special needs• Features adequate for intended use

– Image Quality vs. compact size trade-off• Pocket size >> photo captured

– Output Format: • +JPG High quality? • +TIF? • +RAW?

Summary Cont’d

• Big Sensor + Big lens + computer power > Highest image quality– Big camera, Big $

• Small Sensor + Small lens + limited computer power > Pocket size convenience , Adequate image quality,– Less $, – Less control over camera output

Summary Continued

• Do research camera choices:– Phil Askey: extensive image quality review– Steve’s Digicam: less extensive image quality,

but concise summary with comparison to other camera options. Many more cameras reviewed

– Google: “Camera model” + “review”– Specific camera forums – users own them,

reveal the warts along with the good