Measurement method for image stabilizing systems [6502-22]

Measurement method for image stabilizing systems [6502-22]
Measurement Method for Image Stabilizing Systems
Borys Golika and Dietmar Wuellerb*
a
b
Wahner Str. 1a-3, 50679 Cologne, Germany;
Image Engineering, Augustinusstr. 9d, 50226 Frechen, Germany
ABSTRACT
Image stabilization in digital imaging continuously gains in importance. This fact is responsible for the increasing
interest in the benefits of the stabilizing systems. The existing standards provide neither binding procedures nor
recommendations for the evaluation. This paper describes the development and implementation of a test setup and a test
procedure for qualitative analysis of image stabilizing systems under reproducible, realistic conditions. The basis for
these conditions is provided by the studies of physiological properties of human handshake and the functionality of
modern stabilizing systems.
Keywords: Image Stabilizing Systems, Tremor, Handshake, OIS, VR, Anti-Shake, Shake Reduction
1. INTRODUCTION
In the photographic practice, whether digital or analog, the handshake of the photographer often results in a disturbing
blur. In medicine, the phenomenon of rhythmic, involuntary muscle contractions, occurring in all healthy individuals, is
known as physiological tremor1.
Recently many camera and lens manufacturers have developed various stabilizing systems to compensate for
handshake. The objective of our work was the development of a test method in order to evaluate the quality of image
stabilizing systems. The measurement is based on automated, reproducible mechanical simulation of human handshake
followed by resolution measurement of captured images.
In order to determine the basic test conditions, Bradley J. Davis’ and John O’Conell’s2 method of amplitude
measurement of human physiological tremor was adopted with regard to holding photographic cameras. The ascertained
values were used for the tests.
The “Siemens SFR” (to determine the spatial resolution of digital cameras) is used for blur estimation in this work.
This method was developed in cooperation between Cologne University of Applied Sciences and Image Engineering
Dietmar Wueller, in the context of a diploma thesis by Anke Neumann. MTF describes a contrast decreasing extent in an
output image compared to the test chart. The two methods to measure MTF described in ISO 12233 have several
disadvantages. The visual evaluation method’s results can vary depending on the person performing the measurement.
The Spatial Frequency Response method (SFR) provides moderate results using consumer digital cameras if they have
no access to the raw image data and if the automatic sharpening function can not be disabled3. The “Fit Method” uses
Siemens stars which are periodically sine modulated in the radial direction. Due to the sine modulation the internal
sharpening algorithms of a camera can be avoided. For more information about “Siemens SFR” please refer to a white
paper by Christian Loebich3 and a diploma thesis by Anke Neumann4.
2. IMAGE STABILIZING SYSTEMS
The development of integrated image stabilizers started in 1980’s. Many manufacturers undertook research in this area
and invented various concepts for stabilization, like CCD-Shift or Digital Stabilizers. Handshake detection via integrated
sensors, rapidly moving mechanical elements and digital signal processing with complex algorithms became state-of-theart.
*
Further author information: (Send correspondence to Dietmar Wueller)
Dietmar Wueller: E-Mail: [email protected], Telephone: +49 (0)2234 912141, Fax: +49 (0)2234 912142
Borys Golik: E-mail: [email protected], Telephone: +49 (0)221 8809730
Digital Photography III, edited by Russel A. Martin, Jeffrey M. DiCarlo, Nitin Sampat, Proc. of
SPIE-IS&T Electronic Imaging, SPIE Vol. 6502, 65020O, © 2007 SPIE-IS&T · 0277-786X/07/$18
SPIE-IS&T/ Vol. 6502 65020O-1
2.1. Survey of the Systems
2.1.1. Control Technology
Two angular velocity sensors (gyros) – one for pitch- and one for yaw-axis are used to gather the information about the
shaking of the camera. An additional positioning sensor (e.g. a magnetic hall-effect sensor) detects the current position of
a movable correcting element. Out of the processed sensor data a microprocessor calculates the correction amount and
direction and sends it to the control system. The control system generates the motion parameters for correction element
and runs a voice-coil motor (VCM), piezo-element or another actuator, which is used to move this element.
2.1.2. Optical Image Stabilizer (OIS)
To compensate for handshake, this system, sometimes referred to as opto-electronic image stabilization, uses the optical
path. A movable lens group or a prism with movable surfaces shifts the optical path in order to avoid blurring. The
correction element’s motion is perpendicular to the optical axis in opposite direction to the handshake. The moving force
can be either produced by voice-coils or by piezo-elements.
2.1.3. Electromechanical Image Stabilizer (EMIS)
Electromechanical Image Stabilizing System (EMIS), also referred to as opto-mechanical stabilizer, also uses the optical
path to compensate for handshake. But in contrast to the OIS, the specific feature of this system is imaging chip’s
movement to compensate for handshake. This way it is possible to use any lenses with the camera body equipped with
EMIS. The movement is achieved either by electromagnetic or by piezoelectric actors.
2.1.4. Electronic Image Stabilizer (EIS)
The electronic image stabilization (EIS) does not use the optical path to eliminate the effect of handshake on image
sharpness. Instead of correcting the image on its way to the sensor, this kind of stabilization uses software algorithms to
process the pictures after they have been captured. Sometimes Electronic Image Stabilization is also referred to as Digital
Image Stabilization.
One way to implement EIS is the Advanced Shake Reduction System (ASR) by Samsung. Two shots of a scene are
taken for the stabilization – one blurred picture with the required slower shutter speed to acquire the color and luminance
values and a darker one, using shorter exposure time with sharp edges. These two images are then processed by the
camera software to reconstruct one blur-free picture out of their data.
Some manufacturers refer to electronic image stabilization in their cameras, which should suppress both motion blur
caused by handshake and by moving objects in the scene. The background of these statements is the amplification of the
image signal achieved by increasing the ISO speed. The effect is the shortening of the exposure time, which makes the
influence of handshake almost imperceptible. A faster shutter speed also “freezes” the moving objects in the scene.
Strictly speaking, this method can not be called image stabilization.
2.2. Gyroscopic Sensors
Gyroscopic sensors (gyros), also referred to as angular velocity or angular rate sensors, are inertial sensors. They use the
property of bodies to maintain velocity (linear or angular), unless disturbed by forces or torques as described in
Newton’s law of inertia. The output signal of gyroscopic sensors is proportional to the rate of rotation. It makes them
suitable for detection of rotative handshake motions in image stabilizing systems. The vast majority of stabilizing
systems uses the benefits of such sensors.
Gyroscopic sensors have their origins in mechanical spinning mass gyroscopes, often used in aerospace applications.
An essential part of these gyroscopes is a rotor on an axle which, once spinning, tends to maintain its position in space if
the outside gimbals change. Since the invention of Micro Electro-Mechanical Systems (MEMS) and lithographic
technologies, it is possible to miniaturize gyros and make them affordable. Another important reason for miniaturizing
gyros is the vibrating gyros technology. No bearings are needed to support the mechanics because the rotor is replaced
by a vibrating element. In general, a distinction is drawn between optical and mechanical gyros, whereas only
mechanical ones can be miniaturized. That makes them interesting for consumer electronic applications such as image
stabilizing systems in photographic cameras, camcorders or mobile phones. Various MEMS gyros architectures are
available, using quartz, silicon or piezo-electrical ceramic for the vibrating resonator. The advantage of silicon is that it is
more suitable for the Integrated Circuit (IC) Technology and the resonators are smaller than quartz ones.
SPIE-IS&T/ Vol. 6502 65020O-2
Two separate single-axis gyros are mostly used to detect camera shake. In fact there are gyroscope ICs able to measure
rotation of about up to three axes. Admittedly they are too expensive for consumer electronic applications.
The output of a gyro is measured in millivolts per degree per second (mV/deg/s). The sensitivity of MEMS gyros
varies according to application range. For example the single-axis sensor used in Nikon Coolpix 8800 detects motion in a
range of 0.1° to 1500° per second and outputs 0.66 mV/deg/s5. The power consumption of gyroscopes used in consumer
electronics is lower than 10mW. They cost less then $10, having worldwide annual quantities of over 1 million pieces.
Compound annual growth rate is about 15% (according to Yole Development & Wicht Technologie Consulting/Nexus)6.
3. PHYSIOLOGICAL TREMOR
All humans, not only those with various diseases but even absolutely healthy individuals tremble more or less under
certain circumstances1. In medicine this phenomenon is also known as tremor, which is the most common movement
disorder. Tremor is a rhythmic, involuntary, oscillatory movement of body parts7. It can occur in isolation or as a part of
a clinical syndrome. Tremor comes into being when muscles contract and relax repetitively. Involved body parts are
usually hands, lower arms and head. There are more than 10 various pathological tremors8. Probably the best known
tremors are symptoms of Parkinson’s disease or multiple sclerosis. Uncontrollable shaking movements mark these
illnesses.
Healthy people also exhibit a so-called normal physiological tremor which is not pathological in its nature. Most
people are unaware of this phenomenon, because it is usually not visible7. This kind of tremor affects both men and
women regardless of their age. Physiological tremor can be classified as an action postural tremor which means that it
occurs in action, while a limb (e.g. an arm) is maintaining position against gravity (e.g. holding a photographic camera).
Usually physiological tremor is not a problem and can’t even be seen by the naked eye. But it can be exacerbated by
some factors. First there are some medications (e.g. anti-depressants or anti-psychotics) which can intensify tremble
activity. Some stimulants and toxins like caffeine also have similar effects. And finally there are physiological (e.g.
narcotic or alcohol withdrawal, hypoglycemia) and emotional (e.g. excitement or fear) states, which have a negative
impact on tremble as well.
3.1. Amplitude
Physiological tremor comes into being due to various factors such as mechanical-reflex system and external disturbance2.
Many attempts have already been undertaken to measure tremor amplitude. There have been studies using
accelerometers, digitizing tablets, methods that mimic micro surgical techniques and laser-based systems2. Many studies
focused on single joints, such as the wrist, or did not study the case of a mechanical load held against gravity like
holding a photographic camera. This fact makes a measurement necessary, which has been adapted to a specific tremor
characteristic while taking photographs.
Our objective was to measure the amplitude of physiological tremor in the upper limbs exhibited by healthy people
holding a camera in their hands. The methods should be as simple and comprehensible as possible. Six healthy people
aged 17-35 years were studied. None of the subjects had visible pathologic tremor symptoms. No subjects were taking
any medication known to suppress or exaggerate tremor. A laser penlight weighting 35g was used for this experiment. A
DIN A3 landscape formatted target, consisting of a grid with 1x1 cm squares was used. In the middle of the target, a
cross rule with millimeter-steps was used for further evaluation of the laser light path. The width of the laser light at the
target was about 1 cm at the distance used in the study. The circles on the test chart correspond to the 0.2° to 1.0°
deflection.
Two digital photographic cameras were used to observe the tremor amplitude under three different conditions. A
Nikon D2X with AF-S Nikkor 17-55mm 1:2,8 G ED represented a heavy DSLR with an average size lens. In this case
the camera is held with both hands and additionally stabilized by the head of the test person looking through. Nikon
Coolpix 8400 was representative of a viewfinder camera and was used in two ways, aiming at the target through the
viewfinder and using the LC-Display. In both these cases, the camera was held with one hand only. The difference
between them was the additional stabilization by the head while using the viewfinder. Another digital viewfinder camera,
HP Photosmart R927, was used to take photographs of the laser light path on the test target. Test subjects stood 10m
from the target – far enough to achieve desired accuracy of measurements. They held the cameras either with both hands
(DSLR) or with their dominant hand (consumer compact viewfinder camera), aiming at the center of the test target. The
path of the laser light was captured with the HP Photosmart R927 fixed on a tripod. For measuring the maximum
SPIE-IS&T/ Vol. 6502 65020O-3
amplitude, an exposure time of 5s was chosen to integrate the light path deflection over a longer time period. Each
person was tested under the three different conditions described above. The three trials were spaced at intervals of 3
minutes.
Figure 1. Example of a test image
The images were visually evaluated and the estimated angle values were averaged. After the evaluation, the upper bound
for the required angular travel of the vibrating unit was set to 0.6°.
0.60
0.52
0.50
0.53
0.47
0.43
Amplitude, [deg]
0.52
0.43
0.40
0.30
Yaw
Pitch
0.20
0.10
0.00
DSLR
Consumer camera
(LCD)
Consumer camera
(Viewfinder)
Figure 2. Mean values of the maximum measured amplitudes
The test method described above is generally applicable when measuring the maximum angular deflection holding a
camera in the hand. The estimation of the amplitude of a single tremor oscillation is not possible due to the long time
exposure of five seconds.
3.2. Frequency
The frequency of a camera shake can be derived from the properties of the physiological tremor. The medical literature1
claims that the peak of the tremor activity is at about 8-12Hz. The measurements of the engineers by Panasonic9 showed,
SPIE-IS&T/ Vol. 6502 65020O-4
that the frequency and amplitude are coherent. The higher the frequency of the handshake, the smaller the amplitude
(Figure 3). This means that low and middle frequencies contribute more to the amount of handshake which results in
blurred images. The researches by Ricoh10 leaded to similar results. A diagram in a patent by Ricoh (Figure 4) shows a
high-frequency oscillation (about 10Hz) overlaid by a low-frequency component (about 1-2Hz). This low-frequency
component has much greater amplitude. This “1/f” characteristic of the handshake frequency has an effect on the choice
of proper parameters for a test equipment.
0
5
10
20
15
30
25
Frequency(Hz)
Figure 3. Analysis of image fluctuation frequency9
ROTATIONAL DISPLACEMENT EXAMPLE
TO THE CAMERA AXIS
0.002
0.001
—YAW DIRECTION
0
-0.001
-0.002
-0.003
-----PITCH DIRECTION
-0.004
-0.005
0
0.2
0.4
0.6
0.8
1.0
MEASURED TIME (s)
Figure 4. Angular camera displacement due to handshake10
4. TEST BENCH
Due to the considerations of the characteristics of the tremor and the functionality of image stabilizing systems a
prototype of a test bench was designed. The vibration unit is designed for the reproducible simulation of human
handshake. In connection with the measurement of the resolution of captured images, it is used to analyze the quality of
stabilizing systems.
4.1. Specifications
The device shakes the camera in two defined directions (pitch and yaw) with user-defined frequency and amplitude. It is
possible either to simulate handshake about two axes simultaneously or to use only one single motion direction. The
controlling and parameter inputs are effected by the user via computer.
The vibration unit operates within the frequency range of 0...15Hz and is able to achieve angular motion amplitude of
more than 1°. It suits both light weighted and heavy DSLR cameras and provides enough space to mount any kind. It is
possible to adjust the mounted camera to match its center of gravity in order to achieve reproducible motion.
SPIE-IS&T/ Vol. 6502 65020O-5
4.2. Realization
&
.—
a—
The vibration unit consists in general of two frames nested in each other. This construction allows the simulation of a
handshake about the x- and y-axis (pitch and yaw). The frames are actuated independently. Due to this fact a variety of
motions can be simulated.
Figure 5. Vibration Unit
A closed loop control allows to actuate the frames very precisely due to high resolution position sensors. The additional
advantage of using a closed loop circuit is the stability of the system. The chosen actuators are not limited in their
precision. Figure 6 illustrates the stability of the motions. The horizontal offset of the curves is conditional on the
characteristics of the feedback position sensors. It is of no importance for our measurements.
-0.1
-0.2
-0.3
Angle [deg]
-0.4
-0.5
Yaw
Pitch
-0.6
-0.7
-0.8
-0.9
-1.0
-1.1
-1.2
0.0
1.0
2.0
3.0
4.0
5.0
Time [sec]
-0.1
-0.2
Angle [deg]
-0.3
-0.4
-0.5
Yaw
Pitch
-0.6
-0.7
-0.8
-0.9
-1.0
-1.1
-1.2
2.0
3.0
Time [sec]
Figure 6. Path-time diagram of both frames at 10Hz and 0.5°.
SPIE-IS&T/ Vol. 6502 65020O-6
5. TESTS
There are different approaches to realize a test row to get comprehensible results. Currently an MTF curve of a fixed and
unmoved camera is calculated first. This provides a reference for the specific camera and lens combination. A
measurement combining both varying amplitudes and frequencies is very time consuming. Therefore two possibilities
are available – to vary the amplitude at a fixed frequency or to set a fixed amplitude and go over different frequencies.
This is done for the switched off stabilizer as well as for every stabilizing modus since there are mostly more than one.
For each image in a test row, independent of which component is being varied, the shutter speed is chosen according to
the shake frequency. A complete single oscillation should be captured, this means e.g. testing at 10Hz requires a shutter
speed of 1/10s. The ISO speed is set to 100 and the focal lengths of tested cameras should be similar.
The following charts show, how the evaluation is performed. Figure 7 illustrates the contrast behavior when the
frequency is varied. The amplitude here was 0.2°. The influence of various amplitudes is shown in Figure 8. The
frequency here has been fixed at 10Hz. Figure 9 shows the contrast values at 300 line pairs per image height. This is a
more clearly arranged presentation of the values from Figure 7.
Stabilization OFF
1,1
1
0,9
Contrast
0,8
Reference
Lower Limit
1 Hz
3 Hz
5 Hz
8 Hz
0,7
0,6
0,5
0,4
10 Hz
13 Hz
0,3
0,2
15 Hz
0,1
0
170
420
670
920
1170
1420
Resolution (line pairs / image height)
Continuous Stabilizing Mode
1,1
1
0,9
Contrast
0,8
Reference
Lower Limit
1 Hz
3 Hz
0,7
0,6
0,5
5 Hz
8 Hz
10 Hz
13 Hz
0,4
0,3
0,2
15 Hz
0,1
0
170
420
670
920
1170
Resolution (line pairs / image height)
Figure 7. MTF curves (variable frequency).
SPIE-IS&T/ Vol. 6502 65020O-7
1420
Stabilization OFF
1
0.9
0.8
Contrast
0.7
Lower Limit
0.6
Reference
0.1°
0.2°
0.3°
0.4°
0.5
0.4
0.3
0.5°
0.2
0.1
0
170
670
1170
Resolution (line pairs / image height)
Continuous Stabilizing Mode
1
0.9
0.8
Contrast
0.7
Lower Limit
Reference
0.1°
0.6
0.5
0.2°
0.3°
0.4°
0.4
0.3
0.5°
0.2
0.1
0
170
670
1170
Resolution (line pairs / image height)
"Shoot Only" Stabilizing Mode
1
0.9
0.8
Contrast
0.7
Lower Limit
Reference
0.1°
0.6
0.5
0.2°
0.3°
0.4°
0.4
0.3
0.5°
0.2
0.1
0
170
670
1170
Resolution (line pairs / image height)
Figure 8. MTF curves (variable amplitude).
SPIE-IS&T/ Vol. 6502 65020O-8
1
0.9
0.8
Contrast
0.7
0.6
Stabilization OFF
Continuous Stabilization
Mode
0.5
0.4
0.3
0.2
0.1
0
5
10
15
Frequency, Hz
Figure 9. Contrast values at 300 line pairs per image height taken from charts in Figure 7 (variable frequency).
6. CONCLUSION AND DISCUSSION
Our aim was to develop a reliable test method for the measurement of efficiency of image stabilizing systems integrated
into digital still cameras or lenses. An analysis of the functioning of recent stabilizing systems was performed. Further
studies and measurements of human physiological tremor were carried out in order to define basic conditions for the
tests. The gathered findings provided a basis for the design of a mechanical device simulating human hand tremor. A
prototype device was constructed.
An improvement of the test method can be achieved when considering the suggestions below. First, more detailed
examinations on different subjects with different camera types (DSLR and compact) should be performed to determine
statistically firm handshake properties. In order to do this, some additional equipment such as an accelerometer would be
necessary. An automatic release of camera’s shutter would decrease the testing time, advancing the work flow. If
connected to the control unit the shutter release can be actuated in the exact moment when the oscillation parameters
(velocity and acceleration) conform to requirements. This would make the test results more comparable. An additional
permanent effort of measuring the shutter delay would be caused by this improvement. This delay time must be
considered when actuating the shutter release button. A graphical user interface (GUI) can be developed to simplify user
parameters input and program updates of the control unit. Full automated tests combining automatic shutter release and
controlling program, simulating different oscillation conditions in series, would be possible. A visualization concept for
the test results can be developed in order to represent the acquired MTF values in one single chart. For example, a 3D
surface chart, representing the dependence of the resolution limit frequency, providing only 10% of the contrast, on
oscillation’s amplitude and frequency.
REFERENCES
1. S. Smaga, “Tremor - Problem-Oriented Diagnosis”, American Family Physician 68, pp. 1545–1552, 2003.
2. B. J. Davis and J. O’Connell, “Shoulder, Elbow and Wrist Components of Physiologic Tremor Amplitudes
Measured Using a Laser Penlight,” European Neurology 43, pp. 152–154, 2000.
3. C. Loebich, “Modulated siemens star method to determine the resolution of digital camera systems,” 19.09.2006
(http://digitalkamera.image-engineering.de/downloads/modulated%20siemens%20star%20method.pdf).
4. A. Neumann, “Methods to measure resolution of digital cameras”, Diploma Thesis at the Department of Imaging
Sciences and Media Technology, Cologne University of Applied Sciences, 2003.
5. NEC TOKIN Corporation, “Piezoelectric Devices - Ceramic Gyro,” 20.04.2006
(http://www.nec-tokin.com/english/product/piezodevice2/ceramicgyro.html).
6. T. Link, “Mikromechanische Inertialsensorik,” 15.08.2006
(http://www.hsg-imit.de/pdfs/MikromechanischeInertialsensorik.pdf).
7. F. Regli and M. Mumenthaler, Basiswissen Neurologie, Georg Thieme Verlag, Stuttgart, 1996.
SPIE-IS&T/ Vol. 6502 65020O-9
8. K. T. Wyne, “A comprehensive review of tremor,” JAAPA - Journal of the American Academy of Physician
Assistants 18, pp. 43–50, 2005.
9. H. Kusaka, Y. Tsuchida, and T. Shimohata, “Control technology for optical image stabilization,” SMPTE Journal
111, pp. 609–615, 2002.
10. T. Kitazawa et al., “United States Patent Application No. 20020163581 – Imaging Apparatus, and Method and
Device for Shake Correction in Imaging Apparatus,” 2002.
SPIE-IS&T/ Vol. 6502 65020O-10
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement