a picture can say more than 1000 words comparing camera...
TRANSCRIPT
A picture can say more than 1000 words – comparing camera images Jürgen Bretschneider, 2015
Allied Vision Profile
Foundation: 1989, Headquarters: Stadtroda (Germany)
Employees: > 300 (2015)
Know how:
– Development and production of high-quality digital cameras for industrial and scientific imaging
Research & Development
– Hardware- and software development of digital cameras for the visible and the infrared spectrum
– Standard cameras and also camera solutions for OEM customers
Production
– Company-owned production sites (cleanroom) in Stadtroda and Osnabrück (Germany), Burnaby BC (Canada)
Service
– Consulting teams for distribution and support, worldwide, 24/5
Quality standards
– Compliance with ISO-9001 and ISO 13485
Measurement setup (Ulbricht sphere)
EMVA 1288 standard unified method of measuring a camera’s imaging performance
Ulbricht sphere for C-/ CS- and F-Mount-cameras with homogeneous light source at 632 nm, 529 nm, 448 nm and 850 nm (NIR)
Camera as „black box“ (only pixel size and exposure time must be known)
All measurements use cameras without protection or filter glass
Calibration of the light source and setup of light intensity initially
> 50 measured points to generate curves for sensitivity, linearity, temporal and spatial noise
EMVA 1288 camera data sheet provides measured and derived EMVA data based on 5 cameras of the same type
R, G, B, NIR
Simplified EMVA 1288 related camera model
Photon Shot Noise
photons
Digital value
Temporal: dark noise, quantization noise Spatial: Dark Signal Non-Uniformity (DSNU) Photo Response Non-Uniformity (PRNU)
Noise
- - -
- - -
-
Signal
Description of characteristic EMVA 1288 values
• Sensitivity threshold: point where signal equals temporal dark noise • Saturation capacity: number of electrons that each pixel of an sensor holds, before reaching non-linear response • Dynamic range (DYN): Ratio of the maximum signal (saturation capacity) to the minimum signal (temporal dark noise) • Maximum signal-to-noise ratio (SNR): Ratio of the maximum signal (saturation capacity) to all noise sources • Noise: Unwanted statistical variations of signal produced by the image sensor (temporal dark noise, photon shot noise and spatial noise, like DSNU and PRNU)
NoiseShot Photon NoiseDark
Capacity Saturationmax
SNR
NoiseDark
Capacity SaturationDYN
Description of characteristic EMVA 1288 values
Dark noise
Sensitivity threshold
Saturation capacity
Saturation point
Maximum signal
Dark noise + Photon Shot
noise
SNR at saturation Slope = Quantum Efficiency
Sensitivity threshold (SNR=1)
Caused by irradiance: - Photon shot noise (temporal)
- PRNU (spatial)
Noise SNR bit
48 db 8
24 db 4
Irradiance independent: - Dark noise (temporal)
- DSNU (spatial)
60 db 10 Gain
EMVA 1288 DATA Prosilica GT2750
Photon transfer method: Noise as a measuring tool (example Manta G-223C)
System gain 1/K: 2.561 [e-/DN] Quantum efficiency: 49.205 [%] Sat. Capacity: 10028 [e-] | 20380 [p~] dark noise: 4.608 [DN] | 11.779 [e-] SNR max: 40.012 [dB] | 6.646[bits] Sensitivity threshold: 20.001 [e-] | 40.811 [p~] Dynamic Range DR: 52.065 [dB] | 8.648 [bits]
measured data: System Gain, QE, Temporal Dark Noise, Saturation Capacity, derived data: Sensitivity Threshold, DNR, SNRmax
Temporal dark noise
System Gain QE
Saturation Capacity
Spatial nonuniformity DSNU and PRNU (MG-223C)
DSNU = 4.422 [DN] | 11.326 [e-] PRNU = 1.192 [%]
Deviation Outliers
Model
Model
Spectrogram Method: analyse spatial nonuniformity DSNU and PRNU (vertically and horizontally)
logarithmic histograms of spatial variations – find outlier (Defect Pixels)
Sensitivity curves of selected sensors QE = f (λ)
Manta G-235B (CMOS IMX174LLJ) Manta G-235C (CMOS IMX174LQJ)
448 nm 529 nm 632 nm
Sensitivity estimation: S~QE/dark noise
529 nm 850 nm (NIR)
EMVA data of selected Allied Vision cameras
0
10
20
30
40
50
60
70
80
Sensitivitythreshold
[e-]
QuantumEfficiency
[% @ 529 nm]
SaturationCapacity
[e- *1000]
SNR max [db] DNR max [db] Pixel area [µm²]
Manta G-235C Manta G-223C Mako G-503C
Mako G-192C Global Shutter Mako G-192C Rolling Shutter Manta G-201C
Manta G-145C
appears brightest?
1
1
1
lowest threshold?
Low-light performance of selected Allied Vision cameras
Title Presentation | dd/mm/yyyy | 12
Camera Manta G-235C IMX 174
Manta G-223C CMV 2000
Mako G-503C MT9P006
Mako G-192C GS EV76C570
Mako G- 192C RS EV76C570
Manta G-201C ICX274
Manta G-145C ICX285
Pixel area 34,34 µm² 30,25 µm² 4,84 µm² 20,25 µm² 20,25 µm² 19,36 µm² 41,60 µm²
camera Sensitivity threshold
Manta G-235C 7,07 e- 11.5 [ p~]
Manta G-223C 20.001 [e-] 40.811 [p~]
Mako G-503C 6.072 [e-] 12.157 [p~]
Mako G-192C Global Shutter
22,8 [e-] 62 [p~]
Mako G-192C Rolling Shutter
9 [e-] 28 [p~]
Manta G-201C 9,5[e-] 30 [p~]
Manta G-145C
22,5 [e-] 56,92[p~]
1 MG-503C – really
so sensitive?
Comparing images at low-light condition
Mako G-192C GS Mako G-192C RS
Mako G-503C Manta G-201C
Manta G-223C Manta G-235C
Manta G-145C
Test setup: • Neon tube
5000K • Kowa LM12SC
12mm 1" F1,8 • Camera @
24dB, 300us, Gamma 0,45
Comparing images at low-light condition
Mako G-192C GS Mako G-192C RS
Mako G-503C Manta G-201C
Manta G-223C Manta G-235C Test setup: • Neon tube
5000K • Kowa LM12SC
12mm 1" F1,8 • Camera @
24dB, 300us, Gamma 0,45
Manta G-145C
1
2
3 4
5 6
7
Low-light performance adapted to photons/µm²
Title Presentation | dd/mm/yyyy | 15 Sensitivity threshold (SNR = 1) will be reached earlier at less photons
camera
light densitiy photons/µm² @ Sensitivity threshold
Manta G-235C 0,33 p/µm²
Manta G-223C 1,35 p/µm²
Mako G-503C 2,51 p/ µm²
Mako G-192C Global Shutter
3,06 p/µm²
Mako G-192C Rolling Shutter
1,38 p/µm²
Manta G-201C 1,55 p/µm²
Manta G-145C 1,37 p/µm²
Low-light performance adapted to photons/µm²
Title Presentation | dd/mm/yyyy | 16 Sensitivity threshold (SNR = 1) will be reached earlier at less photons
camera
light densitiy photons/µm² @ Sensitivity threshold
Manta G-235C 0,33 p/µm²
Manta G-223C 1,35 p/µm²
Mako G-503C 2,51 p/ µm²
Mako G-192C Global Shutter
3,06 p/µm²
Mako G-192C Rolling Shutter
1,38 p/µm²
Manta G-201C 1,55 p/µm²
Manta G-145C 1,37 p/µm²
QE
Bright-light condition – 1. high-speed application (shortest shutter time) and 2. high variation in light (high SNR)
camera Saturation capacity (@529 nm)
Manta G-235C 32700 [e-] | 53030 [p~]
Manta G-223C 10028 [e-] | 20380 [p~]
Mako G-503C 4677,3 [e-] | 9300 [p~]
Mako G-192C Global Shutter
5900 [e-] | 16390 [p~]
Mako G-192C Rolling Shutter
10000 [e-] | 27027 [p~]
Manta G-201C 7200 [e-] | 19650 [p~]
Manta G-145C 16600 [e-] | 45996 [p~]
MG-503C – really appears brightest?
Comparing images at bright-light condition
Mako G-192C GS Mako G-192C RS
Mako G-503C Manta G-201C
Manta G-223C Manta G-235C
Manta G-145C
Test setup: • Neon tube
5000K • Kowa LM12SC
12mm 1" F1,8 • Camera @ 4ms
Gamma 1,0
Application task: need for bright image at shortest possible shutter time, SNR isn´t an issue
Comparing images at bright-light condition
Mako G-192C GS Mako G-192C RS
Mako G-503C
Manta G-201C
Manta G-223C Manta G-235C
Manta G-145C
Test setup: • Neon tube
5000K • Kowa LM12SC
12mm 1" F1,8 • Camera @ 4ms
Gamma 1,0
115
155
87
111 221
143
164
7
6 5
4 3
2 1
Bright-light performance adapted to photons/µm²
camera
light densitiy photons/µm² @ Saturation capacity
Manta G-235C
1572 p/ µm²
Manta G-223C
674 p/ µm²
Mako G-503C 1921 p/ µm²
Mako G-192C GS
809 p/ µm²
Mako G-192C RS
1315p/µm²
Manta G-201C
1015 p/µm²
Manta G-145C
1105 p/µm²
Comparing images at bright-light condition
12.5 ms/ 0 db gain/ Gamma 0.45 8.5 ms/ 0 db gain/ Gamma 0.45
Test setup: • Neon tube 5000K • Kowa LM12SC 12mm 1" F1,8 • images adjusted to 250 DN
Application task: high variation in lighting condition, high dynamic scenes, high SNR is needed
Manta G-235C Mako G-503C
115
225
80
195
What about Dynamic at low illuminated image parts?
Title Presentation | dd/mm/yyyy | 22
Manta G-235C Mako G-192C GS Mako G-192C RS
SNR max 45 dB (7.5 bits) SNR max 38 dB (6.3bits) SNR max 40 dB (6.6 bits)
Signal-to-Noise Ratio (SNR)
39 31
19
Noise!
CCD artifacts – blooming and smear
Blooming: Charge overflow (> full well capacity) between neighboring pixels Corrective action: reduction of
the incoming light
Smear: During readout photons generate a charge in the vertical shift register Corrective action : increase of the exposure time, use of a mechanical or LCD shutter, use of flash illumination
blooming and smear – test images of CCD and CMOS sensors
Mako G-192C (EV76C570) Mako G-503C (MT9P006)
Manta G-145C (ICX 285) Manta G-201C (ICX 274) Manta G-223C (CMV2000)
Manta G-235C (IMX 174)
Global Shutter efficiency – unwanted sensitivity
Shutter efficiency = (1 - parasitic light sensitivity)
Define how much the stored pixel value is distorted by incoming light during readout phase of the sensor when the shutter is closed
Former CMOS global shutter sensors are affected by higher parasitic light sensitivity reduces the ability to use very short integration times to freeze motion
E.g. datasheet of the CMOS sensor IBIS5-A-1300 specifies a parasitic sensitivity < 1% up to 400µs at 40 ms readout time
In case of long exposure times, like 20 ms, parasitic sensitivity 400 µs is negligible, but in case of 2µs shutter time this leads to a strong vertical increase in brightness
Nowadays CMOS global shutter sensors provide ratio of 1:2000 or better shutter efficiency
IBIS5-A-1300:
Global Shutter efficiency of selected sensors
Mako G-223B (CMV 2000) shutter time: 1µs
Marlin F-131B (IBIS5-A-1300) shutter time: 2µs
Mako G-030C (CMV300) shutter time: 83 µs
Manta G-145B (ICX 285AL) shutter time: 20 µs
parasitic sensitivity up to 400 µs! (readout time: 40 ms)
Meaning of measured values: Background and pitfalls
Quantum Efficiency QE = f (λ) - Measured for monochrome cameras only at 529 nm and for color cameras at 632 nm, 529 nm, 448 nm - Please use sensitivity curve to estimate how the camera performs at other wavelengths
Signal response vs. sensitivity: - Grey value generated at a certain exposure time has nothing to do with sensitivity - For purposes of comparison you have to consider Photon Shot Noise and Dark Noise
Saturation capacity vs. sensitivity: - The lower the achievable saturation capacity, the earlier the maximum output level is reached - This does not mean, that a low saturation capacity implies a high sensitivity - Comparing Sensitivity threshold and saturation capacity of different cameras you have to consider Pixel area for
the incoming light - light density photons/µm²
Gain: - Higher gain does not mean increased sensitivity, since signal and noise are increased in the same ratio - Applying gain reduces the achievable saturation capacity of a camera - In case of sufficient image quality it is used to shorten the shutter time to prevent motion blur
Tolerances: - Due to manufacturing tolerance of the sensor and camera electronics a real camera may differs from the EMVA1288 test results. The uncertainty in measurement of the QE values is up to 10%
EMVA1288 results and real application demands
You always have to reconsider the optical set-up (lens, filters) and suitable Lighting - Consider Quantum Efficiency is a function of the wavelength
Low-light applications: - Camera qualified with low Sensitivity threshold and high QE - Compare light densitiy photons/µm² @ Sensitivity threshold - Sensors with low temporal dark noise allows more gain without loosing to much
image quality at shorter shutter time
Bright-light applications:
a) High-speed application - Need for a bright image at shortest possible shutter time, (SNR isn´t an issue, 4- 6 bit sufficent) camera is qualified with Low value of light densitiy photons/µm² @ Saturation capacity
- Higher QE means less illumination or shorter shutter speed
b) High variation in lighting condition: - High SNR and large saturation capacity are needed for dynamic scenes with
different lighting situations
Thank You/Q&A
Allied Vision Technologies GmbH Taschenweg 2a 07646 Stadtroda Germany
www.alliedvision.com