OPTICAL IMAGE STABILIZATION - MYTHS & MISUNDERSTANDINGS


OPTICAL IMAGE STABILIZATION - MYTHS & MISUNDERSTANDINGS



OPTICAL IMAGE STABILIZATION - MYTHS & MISUNDERSTANDINGS






An article published in "Amateur Photographer" magazine (23JAN2010 "Stabilisation Systems" pp81-85, 'On Test' - Angela Nicholson, Technical Editor), described results of a comparison test that indicated the marked superiority of Canon's lens-based IS Optical Image Stabilization system.

The purpose of this forum article is to analyse the methodology and findings, reveal misconceptions of optical engineering principles, and demonstrate experimental bias.

Before commenting on the AP article I want to describe optical image stabilization (OIS) as it applies to DSLR cameras. Canon & Nikon make camera lenses that incorporate OIS. Canon's trade name for OIS is "IS" & Nikon's, "VR" (Vibration Reduction). Sony and a few other DSLR manufacturers incorporate OIS within the sensor cradle. Sony's trade name for their sensor-based OIS is "Steady Shot", a technology developed by Minolta and featured in the Konica-Minolta Maxxum 7D in 2004 as "Anti Shake".

Why do Canon & Nikon not have sensor-based OIS? The reason is fairly straightforward. Canon's first OIS system was incorporated into their EF75-300/4.5-5.6IS USM zoom in 1995. The lens was intended for use with their range of EOS SLR's. It would have been impractical to move the film chamber, so a lens-based OIS system was their only feasible option. Nikon followed somewhat belatedly in 2000 with their 80-400mm f/4.5-5.6D, intended primarily for use with their F5 SLR.

I want to emphasise from the outset that neither Canon's or Nikon's lens-based OIS or Sony's sensor based OIS are ideal. There are fundamental opto-mechanical reasons why total compensation for hand held camera shake is beyond the capabilities of either technology.

When a photographer is composing a picture in a DSLR optical viewfinder the image seen is subject to four induced movements:

Roll	-  rotation about the camera lens optical axis
Pitch - rotation in the vertical plane of the optical axis
Yaw - rotation in the horizontal plane of the optical axis
Piston - axial shift along the optical axis

Optical Image stabilization systems
The lens-based OIS designed by Canon initially compensated for pitch & yaw. Their latest OIS can in a few specific lenses also compensate for either roll (limited rotation - angle movement in pitch but also simultaneous side to side), pitch & yaw (Hybrid IS), or pitch, yaw & piston.

The sensor-based OIS designed by Konica-Minolta can only compensate for pitch & yaw, but it is possible to also compensate for a limited roll movement.

Angled camera movements that cause the lens to tilt, tilts the image plane. Ideally the sensor should also tilt with the image plane. .

So how does lens-based and sensor-based OIS work? What is the underlying technology?



SENSOR-BASED OIS



SONY STEADY SHOT SENSOR & CRADLE SONY STEADY SHOT SENSOR & CRADLE ASSY

http://www.sony.net/SonyInfo/technology/technology/theme/alpha_01.html


The Konica-Minolta sensor shift OIS comprises two elements. A means of detecting jerky movements of the camera, and a means of sliding the sensor laterally across the focal plane of the lens. Sudden accelerations of the camera are detected using a pair of gyro-accelerometers (pitch & yaw), and sensor movement using a pair of Hall sensors either side of a magnet. The magnet is located on the carriage that holds the cmos sensor and the Hall sensors are located on the support frame. Two piezo-actuators, a pitch actuator located vertically on the cmos sensor carriage, and a yaw actuator located horizontally along the lower edge of the support frame, make compensatory movements in x-y co-ordinates. (The x,y,z, co-ordinate system is defined as z - along the optical axis; x - horizontal orthogonal plane; y - vertical orthogonal plane).
CRADLE GUIDEWAYS SONY PIEZE ACTUATORS
Camera accelerations are resolved into pitch (y-axis) and yaw (x-axis) movements. The extent to which the cmos sensor carriage is shifted in x & y is controlled by a PIC that requires focal distance information from the lens. The maximum shift is dependent on the length of the piezo-actuator, and because an Fx cmos sensor DSLR body provides more space it can accommodate longer piezo-actuators. Consequently an Fx sensor-based OIS can compensate for pitch & yaw camera movements to a greater extent than an APS-C sensor-based OIS. The longer piezo-actuators used in an Fx sensor-based OIS are also capable of moving a greater mass. In the case of the Sony alpha 900 DSLR OIS, 50% more mass. The cmos sensor and carriage however have proportionally less mass than the APS-C equivalent, and the compensatory movements are effected more rapidly. The linear travel, being longer, can also compensate for image shifts of proportionally longer focal length lenses, greater than the 1.5x crop factor.

What the Konica-Minolta sensor-based OIS cannot do at present is compensate for roll movements.

The advantages of sensor-based OIS are three-fold.


a) It can compensate for pitch & yaw movements using any lens, providing that lens sends focal distance information to the OIS PIC. The PIC has to have focal length and focussing distance information to calculate the precise x - y compensatory movements of the piezo-actuators. If a lens is fitted that can only be used in manual mode, this information is not provided. The camera does not even recognize that a lens has been fitted. The PIC defaults to a focal distance set to infinity, and a standard focal length, corresponding to the diagonal of the cmos sensor (43mm in the case of an Fx cmos sensor). If say a manual 300mm f/2.8 telephoto lens is fitted, the compensatory movements of the piezo-actuators will be less than necessary, by a factor of ≈ 8.

b) Because the OIS is located on the sensor carriage and support frame you only pay for the technology once.

c) The piezo-actuator acceleration parameters can be tailored to the DSLR body mass.

The disadvantages of sensor-based OIS are four-fold.


a) Their is no OIS compensation to the DSLR optical viewfinder (not the case for a DSLR style hybrid camera with a digital viewfinder). This can in certain circumstances cause a framing disadvantage compared to a DSLR with lens-based OIS. However, unless the DSLR optical viewfinder has 100% frame coverage, what framing advantage the lens-based OIS may offer is to a large extent offset.

b) The image plane shift produced by camera movement using a long telephoto lens can exceed the maximum travel of the piezo-actuators.

c) The acceleration parameters are tailored to the DSLR body mass, not the lens + body mass.

d) Because of the small yet finite time the DSLR mirror takes to flip up, a sensor-based OIS cannot compensate for camera shake at the very instant the shutter is opened.

LENS-BASED OIS


http://www.canon.com/camera-museum/tech/room/tebure.html


"Camera shake is a major cause of blurred image especially with telephoto lenses. Normally, a shutter speed at least as fast as the reciprocal of the lens focal length, (1/focal length) sec., can prevent a blurred image due to camera shake. However, under low-light conditions or with slow film, a slower shutter speed will be required, resulting in image blur for handheld shots. Canon started development of IS (Image Stabilizer) technology in 1980s and introduced EF75-300mm f/4-5.6 IS USM in 1995, the world's first interchangeable lens for 35mm SLR with a built-in image stabilizer. The vibration detecting gyro sensor detects the level of camera shake and the actuator moves a part of the optical system (IS lens group) vertically to the optical axis depending on the degree of camera shake, to stabilize the image on the film plane. Normal image stabilizing is suitable for shooting of stationary subject. However, EF300mm f/4 L IS USM has "IS Mode 2" to stabilize finder images at panning shots of moving subjects."
miniaturised Canon OIS CANON OIS lens yoke assy

http://www.canon.com/camera-museum/tech/report/200802/report.html

CANON 3-accelerometer OIS

http://en.wikipedia.org/wiki/File:Canon_EF_IMG_0067.JPG


The Canon lens shift OIS comprises three elements. A means of detecting jerky camera movements within the lens and a means of sliding a lens element orthogonally to the optical axis in x & y. Sudden accelerations of the camera are detected using either two (pitch & yaw) or three (roll, pitch & yaw) gyro-accelerometers mounted within the lens mount, and OIS lens movement using a pair of magnets mounted perpendicularly to one another attached to the OIS lens yoke, and a pair of in-line coils mounted within the OIS support frame. The coils are also used to shift the OIS lens yoke in x & y, against a pair of springs and dampers. A third spring restrains the OIS lens yoke along the z-axis within the support frame, and ensures the OIS lens element maintains orthogonality to the rear lens node.

Camera accelerations are resolved into pitch (y-axis) and yaw (x-axis) movements. The extent to which the OIS lens yoke is shifted in x & y is controlled by a PIC on the lens control circuit board that is supplied focal distance information. The maximum shift is dependent on the length of the coils and magnets, and the space available near the rear node of the lens system within which the yoke can move.

Canon lens-based OIS developments resolve limited roll movements into equivalent x & y translations. An additional coil-magnet actuator also shifts the OIS lens yoke along the z-axis to compensate for piston movements.

The advantages of lens-based OIS are five-fold.


a) Because the OIS lens yoke is a lot less massive than an APS-C or Fx sensor-based OIS carriage and support frame, it can be accelerated and decelerated a lot more rapidly, and can therefore compensate for more violent camera movements.

b) Because each OIS lens has a dedicated OIS lens yoke, the compensation can be tailored to suit the lens + body mass.

c) The OIS lens yoke can translate the image in x & y over whatever distance is required to compensate for camera movement. Image shifts that could not feasibly be accommodated by a sensor-shift based OIS are feasible with a lens-based OIS.

d) Because the image stabilization is effected within the lens, the viewfinder optical image is also stabilized. This can offer a framing advantage, especially when panning. (Canon & Nikon OIS can be restricted to yaw only when panning).

e) Because the OIS is lens-based, image stabilization is effected at the instant the shutter is opened.

The disadvantages of lens-based OIS are four-fold.


a) The technology has to be fitted to each lens and has to be paid for time and time again. The cost of lens-based OIS is far higher than sensor-based OIS. The Canon EF70-200F4L IS USM lens used in the AP test costs £400 more than its non IS sibling. The Nikon 55-200F4-5.6G AF-S DX VR costs £60 more than its non VR sibling. When you compare the prices of Canon IS & their non IS siblings these price differences are repeated, e.g. EF 100F2.8 IS USM macro - £345; EF 70-200F2.8L IS USM - £350.

b) There is no OIS when a manual lens is fitted. An obvious point you might think. Neither Canon or Nikon make image stabilized fixed focus wide angle, standard or short telephoto lenses. (24mm thru 135mm). If OIS is of benefit to a compact mid-range zoom, for instance the Canon EF-S 28-135F3.5-5.6 IS USM, why would it not be equally beneficial to their EF 24F2.8; EF 50F1.4 USM; EF 85F1.8 USM & EF 135F2L USM lenses? When you use a Sony alpha 900 with 50F1.4AF lens, its image stabilized. There is a low light shooting advantage that no Canikon DSLR can ever compete with because there are no Canon or Nikon image stabilized standard lenses! And ironically the type of camera movements that occur when a short lens is fitted are the type of movements that a sensor-based OIS compensates best of all.

c) Because of miniaturisation constraints lens-based OIS can be impractical for ultra-wide angle lenses. It is argued that OIS is not needed for ultra-wide angle lenses, but some of the new ulltra-wide to wide angle zoom lenses intended for use with Fx format DSLR's are long, bulky and heavy, and benefit from OIS.

d) Optical viewfinder image stabilization can in certain circumstances produce nausea. (Nikon have introduced an OIS option to their VR system so that it only operates the instant the shutter is opened). I find this effect questionable - see Myths & Misunderstandings (g).

Comment


You have to seriously ask yourself the question, is lens-based OIS worth the price differential? Given that so many useful lenses in the Canikon stable are not image stabilized, and the cost of their OIS compared to non OIS equivalents, is the advantage this technology confers yet another illustration of the law of diminishing returns? It seems to me that sensor-based OIS is far more practical and economic.

Myths & Misunderstandings


Having described the two technologies and how they function, I will now deal with the myths and misunderstandings that have become inculcated in the minds of photographers, professional and amateur alike.

a) Sensor-based OIS works with any lens.


I have seen this statement on numerous photography and camera review websites. Too numerous to list. This is the OIS myth of all OIS myths. Where the idea comes from baffles me. It does however reveal a complete lack of understanding as to how the pitch and yaw compensations to image shift are performed.

Pitching & yawing the camera produces an angular shift on the image plane (sensor). This has to be converted into x & y-axis translations. In order to do this the image shift angle has to be resolved into an x-axis & y-axis angle shift, and then into a linear x-axis & y-axis compensatory movement.

In order to convert the angular image plane shift to a linear displacement the prime focal length of the lens and its focal distance has to be fed to the OIS PIC. The lens has to provide this information. It provides it by sending two pieces of data, the prime focal length and the AF focus setting. From these two pieces of data the PIC calculates the back focal length and then uses it to convert the angular shift to a linear displacement using a simple trigonometric relationship.

In the absence of focal length and focussing distance information the PIC defaults to a standard focal length (the sensor diagonal length) and infinity focus. Sensor-based OIS in these circumstances will over-compensate for wide angle lenses and under-compensate for telephoto lenses. What offsets the under/over-compensation is the extent of the image shift for a given angular movement. The same angular movement using a wide angle lens will produce a much smaller image shift than that produced by a long telephoto lens.

b) Lens-based OIS offers more compensation in terms of f-stops than sensor-based OIS.


This myth is based on a combination of largely untested marketing claims, and the commonly held belief that the lower mass OIS lens yoke and carrier assembly used in lens-based OIS can be accelerated and decelerated faster than the sensor carriage and carrier in sensor-based OIS.

To some extent the notion holds good, but only to a point. It depends, to a far greater extent than seems to be appreciated, on the focal length of the lens, and the combined mass of lens & DSLR body, relative to the photographer's body mass.

Lens vs sensor-based OIS was the subject of the AP23JAN2010 article which I deal with at the end of this article.

c) OIS is not needed when shooting ultra-wide angle landscapes.


Working from the presumption that small angular camera movements produce equally small if not undetectable angular image shifts when using an ultra-wide angle lens focussed at or near infinity, it is argued that OIS is unnecessary. There are two factors not being considered in this presumption. Firstly that the lens focal length is all that matters; it isn't, it is the back focal length that matters, and all ultra-wide angle lenses designed for DSLR's are retro-focus. A 16mm ultra-wide angle DSLR lens designed for an Fx format sensor has a back focal length about 3 times the prime focal length. If OIS is effective when using a standard 50mm focal length lens it will be just as effective using a 16mm ultra-wide angle lens.

Some of the new Zeiss and Sony G series aspheric apochromatic ultra-wide to wide angle zooms are bulky and heavy. Their use benefits from sensor-based OIS.

The principal source of this myth arose from Canikon marketting hype. Until Spring 2010, it was not feasible to fit OIS in ultra-wide angle lenses because of space restrictions. In order to downplay the disadvantage conferred to Sony's sensor-based OIS, they argued it was not needed in ultra-wide angle lenses. A self-serving argument worthy of a politician. Needless to say however that Canon answered the restriction by successfully miniaturising their OIS so it can be accommodated within APS-C and Fx format ultra-wide angle zoom lenses, and Nikon have succeeded in doing the same. What neither have been able to do as of March 2010, is shrink their OIS so it can be installed in their fixed focus ultra-wide angle lenses.

d) Only lens-based OIS can compensate for the camera being angled right or left and up or down.


When I first came across this misguided notion I almost choked on my Cornflakes. I was reading Angela Nicholson's 'On Test' article, "Image Stabilisation" in AP23JAN2010_pp81-85. I have scanned the article for reference_ [AP23JAN2010_pp81-85_3.4Mb zip].

This is what she wrote: "Camera shake can result in the camera moving in several directions, and often in several directions at once, but not all can be corrected. Most lens and sensor-based systems correct for horizontal and vertical movement (with diagonal movement being a combination of these two). Horizontal and vertical movements are also often accompanied by the camera being angled left or right and up or down, which can only be corrected by lens-based systems".........

How is it possible that a presumably experienced photographer can fail to comprehend that angular camera movements translate into x & y image shifts, just as simple x & y sideways movements also translate into x & y image shifts?

What does she think happens when the camera is angled downwards (pitch) or sideways (yaw) that produces a distinct movement incapable of being translated into a corresponding x & y-axis displacement?

Lets go through her statement and dissect it in an attempt to hopefully arrive at some understanding of her thinking on the matter:

"Camera shake can result in the camera moving in several directions, and often in several directions at once, but not all can be corrected"......

I would dearly love to know how any object can move in more than one direction at once! I think I understand what she is getting at. If we assume camera movements can be resolved into separate horizontal and vertical movements, then any diagonal movement will be a combination of both. But no engineer I've ever come across thinks that way. Any object (be it a Newtonian particle or a house brick) can only move in a single direction at any instant. This is one of the principles behind Newton's second law of motion. It can change direction, but only when acted upon by an external force.

"Most lens and sensor-based systems correct for horizontal and vertical movement (with diagonal movement being a combination of these two)"........

This part of her statement conflates her opening remark and justifies my conclusion that she thinks of camera movement as being the resolved vectors in the x & y-axes rather than the vector itself. (If you remember your high school physics and mathematics, a vector has direction and magnitude).

"Horizontal and vertical movements are also often accompanied by the camera being angled left or right and up or down, which can only be corrected by lens-based systems".........

Here we have the nub of her argument and the statement that reveals a complete misunderstanding of the physics describing what happens when a hand held camera shakes. She is confusing horizontal and vertical camera movements with angular camera movements up & down and side to side (pitch & yaw). In terms of what happens to the image on the image plane, if the camera is simply translated sideways in x & y, if the subject lies at infinity, there is no image plane shift, because the angular shift is zero. The closer the subject lies along the optical axis (z-axis), the greater the angular shift, corresponding to a parallactic displacement across the image plane. Pitch & yaw also produce horizontal and vertical shifts in the image plane. Except for a subject at infinity, there is no distinction.

The confusion probably arises from a Canon press release on 22JUL2009 announcing their "Hybrid IS" anti-roll feature. Canon "Hybrid IS" uses an additional angular velocity sensor to detect the extent of angle-based camera shake. All this does is feed angle based movements directly to the OIS PIC rather than have the PIC reduce them to x-y translations beforehand. It offers a marginal advantage when shooting macro with a short telephoto, and purportedly offers a distinct advantage when using a long telephoto which even when held near the centre of gravity has a tendency to swivel about due its momentum.

After repeated requests for an explanation I eventually received a quite fulsome reply from AP's Editor, Damien Demholder:

"The benefit of an in-lens system is that it can be developed especially for the type of shake the focal length, physical length and the weight of a particular optical construction is likely to produce. Lenses of light, short barrel tend to suffer an up/down motion parallel to the subject plane, while longer, heavier models incur motion that angles the film plane against the subject - the barrel-end waves up and down to a greater degree than the film height moves. In longer lenses this angled motion can be compensated for in a way that is, at the moment at least, not possible in a sensor based stabilisation system. I asked Minolta to develop it in the Dimage days, and I remind the same people now I see them in Sony uniform - but nothing yet. This type of sensor movement - that would see the assembly angling away from the camera back - would also be useful for the type of conditions in which we favour a technical camera with free moving film and lens panels. At present sensor systems can only shift in planes parallel to the camera back - which of course does nothing to correct angled movement."

Whether the camera is simply translated horizontally or vertically or a diagonal combination thereof, or tilted vertically or horizontally (pitch & yaw), all any OIS system can do is translate the concomitant angular image plane shift into an x & y-axis displacement. Both sensor and lens-based OIS systems accomplish this in the manner I have already described. (Ref: Footnotes)

e) Sensor-based OIS can only compensate horizontal and vertical movements.


This is a corollary of myth (d). It arises from the knowledge that a sensor-based OIS shifts the sensor about the x & y-axes. This knowledge is then conflated with the misguided notion that a lens-based OIS somehow compensates for camera tilt by tilting the image plane. This is complete and utter nonsense. It is a classic schoolboy physics howler.

You can test this for yourself. Take a positive lens (a simple magnifying glass will serve), and project an image onto a room wall of a daylight illuminated window. Hold the lens square to both wall and window. Now tilt the lens. Note the image does not shift. When you tilt a lens the optical axis does not tilt with it. The same applies to the negative or concave lens used in a lens-based OIS.

Lens-based OIS works by translating the OIS lens yoke about the rear node, to compensate for pitch & yaw. The OIS lens yoke is moved not only up & down and from side to side, and any combination thereof, but in such a way that the centre of the lens maintains a constant distance from the rear node, and orthogonal to it.

f) Only lens-based OIS offers any means of compensating for roll.


Prior to Spring 2010 no camera OIS system compensated for roll. During the summer 2009 Canon announced roll compensation in some of its forthcoming mid-range telephoto lenses. The plain fact of the matter is no OIS system can compensate for roll in the strict sense of the term. If you were to place the DSLR on a turntable with the lens optical axis in line with the turntable rotation axis would you expect the OIS to de-rotate the image plane?

Canon is being intentionally misleading in my opinion. No lens-based OIS can de-rotate an image. What they have done is rewrite the translation algorithm to accommodate a limited roll movement using a conversion matrix that translates roll into equivalent pitch & yaw movements. It is approximate and will work only for limited roll movements.

The same sort of thing could be accomplished by a sensor-based OIS, but it isn't, and there is a good reason why Sony have not bothered with it. If you stop to think about what is involved in converting limited roll movements into equivalent pitch & yaw movements, the same result can be achieved using pitch & yaw movements alone.

f) OIS provides greater compensation in terms of f-stops with Fx format than APS-C or 4/3 format DSLR's.


This myth arises from comparing apples and oranges. The crop factors of APS-C (x1.6 or x1.5) & 4/3 (x2) sensor formats have a camera movement magnifying effect compared to an Fx format sensor. The reciprocal focal length minimum shutter speed rule of thumb which holds good for Fx format camera lenses from about 24mm to 500mm focal length, is the baseline from which OIS compensation in f-stops is made. The same OIS technology incorporated in an APS-C or 4/3 format DSLR will produce compensation in f-stops inversely proportional to the crop factor.

It is not due to some inherent drawback with APS-C or 4/3 DSLR OIS, it is a consequence of crop factor.

The comparison that ought to be made is between lenses with the same angular field of view. That is lenses whose focal lengths are proportional to the crop factor. So if I take a 200mm telephoto on an Fx format DSLR I should compare OIS compensation with a 125mm telephoto on an APS-C DSLR and 100mm telephoto on a 4/3 DSLR.

When a direct comparison is made on this basis, there is no inherent advantage to the Fx format DSLR OIS.

g) Optical viewfinder stabilization with lens-based OIS makes you sea-sick.


Optical viewfinder stabilization has been a feature of movie cameras for decades. It may be comparatively recent for DSLR's, but why is it that movie makers don't complain of feeling nauseous when panning and zooming shots with a stabilized camera? Are we to take it that motion picture photographers are a different breed than stills photographers? Unless you pan a lens-based DSLR quickly, and jolt it up and down (pitch) simultaneously, the inertial lag in the optical viewfinder is minimal. Even when you pan quickly you are concentrating on the subject somewhere near the centre of the frame. You won't have time to become aware of the inertial lag in any case. Yet another urban myth.

h) A tripod offers little advantage when using a long OIS telephoto lens.


Try telling this to a sports or wildlife photographer and you'll get howled out-of-court. Just because OIS offers a few f-stops compensation for camera shake, doesn't mean you can get away with hand holding a heavy 300mm f/2.8 lens and expect to get the maximum resolution of image detail. OIS works best for wide angle thru' short telephoto lenses in situations where a slight loss of image detail is of no consequence. If you want to get the sharpest shots possible, use a tripod, and remember to switch the OIS off.

The AP article.


I will leave it to you, the reader, to read and inwardly digest Angela Nicholson's test report [AP23JAN2010_pp81-85_3.4Mb zip]. There is a fundamental problem with the adopted methodology. Basically the combined sensor and lens resolution is evaluated with the OIS off and the OIS on. The average score was determined by taking the mean from 10 frames of the standard AP resolution test chart OIS off & OIS on - making 20 per DSLR body + lens combo, all shot at ISO200. The resolution limit in line pairs per millimeter was converted to line widths per picture height x 1,000, and the increase in resolution expressed in the same system as a difference. (Ref: footnote)

This method would be OK if all the camera + lens combos weighed almost the same, but they don't, they range from 1.676 lbsf (Canon EOS 1000D + EF-S 55-200F4-5.6 IS zoom) to 3.34 lbsf (Canon EOS 5DMkII + EF 70-200F4L IS USM zoom), a difference in inertial mass of 100%. The method would also be fair if the lens + sensor combos had similar resolving power, but they don't, they range from the APS-C format Canon EOS1000D 10.1Mp (6 micron photosite) thru' the Panasonic Lumix DMC-G1 micro-4/3 12.1Mp & Olympus E-620 4/3 13.1Mp (4.3 micron photosite) to the Canon EOS 5DMkII Fx 21Mp (6.4 micron photosite).

The test report doesn't state what relative aperture the test frames were exposed at. I assume the centre of each frame was examined when the resolution was measured. Assuming each lens was set to its maximum relative aperture at 1/15s and a stop less at 1/8s, and that all the lenses are diffraction limited when wide open (a big assumption I know), then the resolving power of the lenses in line pairs per millimeter for a 40:1 contrast black & white test target can be calculated from the Rayleigh Limit formula rho = 1.22 x wavelength x focal ratio.

There is one other caveat. Providing the Nyquist limit of the sensor exceeds the resolving power of the lens there will be no Nyquist limit noise effecting the results. This was the case for most of the camera + lens combos used, but not all where in some instances the f/4 resolving power is cut off towards the Nyquist threshold. The resolution values are tabulated under RHO/NY lppm.

The difficulty in changing the relative aperture between the 1/15s & 1/8s exposures is that in stopping down the lens you are reducing the resolving power. For a diffraction limited lens the resolving power is reduced by a factor of √2. By doing this, when you compare the average resolution ratings, OIS on & off, between the 1/15s & 1/8s exposures, you are not comparing like with like.

I also have a problem with her choice of the Sony alpha 550 amongst the selected group of sensor-based OIS DSLR's. There is no direct comparison between an Fx lens-based OIS and an Fx sensor-based OIS DSLR, and it leads to a very misleading conclusion, which in my opinion betrays a Canikon bias.

Table of results published in "Amateur Photographer" magazine 23JAN2010 "Stabilisation Systems" pp81-85 edited by Chris Lord (I have added four columns on the right; RESOLUTION INCREASE expressed as a percentage; COMBINED MASS in imperial pounds; ENHANCEMENT NORMALIZED & RHO/NY the lens resolution and sensor Nyquist noise limit in line pairs per millimetre.)

EDITED AP TABLE OF RESULTS

1. Shutter speeds equate to +3EV & +4EV approximately.
2. Normalization to the mass of the Sony alpha 900 + Sigma 50-200F4-5.6 DC OS combination. This is not a realistic base comparison because the Sigma DC lens is intended only for APS-C format cameras and the Sony alpha 900 would switch from Fx to APS-C imaging format. It was all that could be done given the test results as presented. The ideal combination would be Sony alpha 900 plus SAL-70200F2.8G a combined inertial mass of 4.828 lbs. The SAL-70200F2.8G also has a resolving power matched to the Exmor 24.6Mp sensor. Sony discontinued their kit Fx zoom lens in August 2008.



I wish to draw your attention to the Canon EOS 5D MkII & EF70-200F4L IS USM and the Sony alpha 550 & Sigma 50-200F4-5.6 DC OS. At 1/8s the increase in resolution values are 10 & 4.4 respectively. Assuming the Canon lens was set to 200mm, the Sigma lens ought to have been set to 125mm. A 35mm equivalent focal length of 200mm is claimed for all the images.

The way the increase in resolution is presented as a difference between line widths per picture height x 1000 does not actually describe the effect of the OIS. The increase ought to have been expressed as a percentage improvement, not a simple difference. This would give values of 57% & 29% respectively. The reason for expressing the improvement in resolution as a percentage increase is because it evaluates the increase in the lens + sensor resolution as a factor that may be compared to each of the other camera body + lens combinations. Since the different camera body sensors and different lenses, each have their own different combined resolving powers, it is the proportional increase in resolution the OIS provides in each case that we are interested in, not simply the difference in resolution.

The implied advantage of the Canon Fx format lens-based OIS needs to be viewed with scepticism. Why did Angela Nicholson choose the Sony alpha 550 and not the alpha 900? The only way to make a meaningful comparison between lens and sensor-based OIS is to use DSLR's with the same sensor format, and lenses having comparable resolving power. Judging from the scores achieved by the Canon EF lens compared to those achieved by the Sigma lens, the Canon 5D EF lens + sensor combination clearly yielded a significantly higher resolution. I put this point to her, this was her explanation:

"On the matter of the inclusion of the EOS 5D Mark II, this was intended to demonstrate that sensor size can have a bearing upon how a stabilisation system performs. It wasn't intended to show that Canon's IS is better than anything Sony has to offer."

When evaluating the benefits of OIS one of the factors that needs to be considered is the inertial mass of the camera body and the lens. Camera shake is induced by uncontrolled muscle movements. These have a particular range of forces for any particular individual. More massive cameras tend to respond to these forces more slowly than lighter cameras (following Newton's second law, force equals mass times acceleration). The Sony alpha 550 & Sigma 50-200F4-5.6 DC OS lens has a mass of 2.25 lbs. The Sony alpha 900 & the same lens would have a mass of 2.8 lbs, a 29% increase. This additional inertial mass equates to a 38% increase in resolution, rather than 29%.

But if I simply left it at that I would be overlooking one other factor, the effect on the Canon's OIS of its inertial mass. The Canon 5D MkII & EF70-200F4L IS USM lens has a mass of 3.34 lbs. The Canon camera & lens is 54% more massive than the Sony alpha 550 and the Sigma lens. This equates to a 31% OIS bias skewed in favour of the Canon camera. Correcting for this bias, the percentage improvement in resolution between the Canon 5D MkII and the Sony alpha 550 would be 27% & 29%. Correcting for the bias introduced by choosing the Sony alpha 550 instead of the more appropriate Sony alpha 900, the results would have been 30% & 31% respectively. Not so pronounced a difference as implied by the 10 & 4.4 resolution difference values! See what I mean by experimental bias?

According to my evaluation of Angela Nicholson's OIS test, the Canon 5D MkII & Sony alpha 900 would have come out neck and neck. Not the conclusion she drew from the way she chose to present her results. "Generally, it seems that lens-based stabilisation systems have a slight edge over the sensor-shifting competition, with the Canon Image Stabilization (IS).....mechanisms being particularly impressive". I leave you to draw your own conclusions.

Having used both of these DSLR's and on the balance of its performance as a stills camera chosen the Sony alpha 900, the extrapolated results tally with my user experience. The Exmor 24.6Mp sensor out-resolves the Canon cmosT4 21 Mp sensor. The Sony SAL-70200F2.8 G out-resolves the Canon EF70-200F4 IS USM lens. The DSLR bodies have similar masses and the lens lengths are similar. If these DSLR & lens combos were compared my money would be on the Sony alpha 900.

How ought the experiment have been conducted?

What you want to know from an OIS test is how many f-stops advantage it provides. What this means essentially is what do you class as an acceptable picture? How low does the resolution have to be degraded before it becomes unacceptable. This obviously depends on what size of picture you intend to print, and at what viewing distance it will be seen. Lets assume for the sake of argument that you want an A4 print and it is to be viewed from the standard distance of 10-inches (unit magnification). An acceptable print viewed at 10-inches needs a print resolution of 300dpi. An A4 print measures 297mm x 210mm. The minimum print resolution must be 3508 x 2480 pixels. All the cameras used in the test had sensor resolutions greater than this, so no problem there.

The highest resolution is going to be obtained with the camera mounted on a tripod with the OIS off. This would have been my baseline comparison. The relative aperture of the lens needs to be held constant, so the lens + sensor resolving power remains fixed. This will then give a level playing field when comparing resolution ratings when the camera is hand held.

What exposure times to use?


Instead of picking a pair of exposure times approximately 3EV & 4EV below the reciprocal focal length rule of thumb (factored for crop factor), I would have picked a wider range, starting at the reciprocal focal length speed and decreasing in 1EV steps to +4EV. I would have maintained a constant relative aperture (say f/4 or f/5.6) by changing the ISO setting, from ISO100 thru' ISO800. All the cmos & mos sensors in the cameras used in the test had acceptable noise levels between ISO100 to ISO800. Increasing the read out gain would not have interfered with the result.

I would also have increased the population sample, and instead of calculating the mean, I would have calculated the population standard deviation of the mean. To do this I would have taken 30 shots with the lens set to f/5.6 say, tripod mounted with OIS off, and then hand held at EV+0; EV+1; EV+2; EV+3 & EV+4. I would not have discarded bad frames, I would use all 30 and calculate the standard deviation of the mean for each sample.

How ought the results have been presented and compared?


The resolution ratings, when reduced to a standard deviation of the mean and the mean itself, should have been compared as a proportional difference, between OIS off and OIS on. I would also have included a second comparison between OIS on and the tripod rating.

The results I would expect from this method is that the enhancement in resolution rating would decrease at each stop increase. I would expect the highest rating with the tripod mounted setup, followed by a steadily decreasing rating for the hand held camera, for each stop. The reason I would expect this result is because no OIS system can follow all pitch and yaw movements exactly in time, there is a slight lag, and so the longer the camera + lens combo is handheld, the worse the compensation becomes. And assuming the photographer is not becoming fatigued, the decrease ought to be proportional to the time the camera is hand held.

I think one of the reasons the AP test results are so inconsistent is because of the methodology adopted. It overlooks differences in camera + lens inertial mass, differences in sensor + lens resolving powers and Nyquist thresholds, differences in lens resolving power in stopping the lens down 1 f-stop between test exposures, and has too small a population sample to be statistically significant. But given the amount of work involved in following a more rigorous procedure, it would be unrealistic to expect a magazine editor to have the time to do it.

FOOTNOTES


Full-frame compared to DX



Eagle-eyed viewers will no doubt have noticed that the average score data at any particular focal length/aperture combination is distinctly higher on the full-frame FX compared to DX. This may at first sight appear unexpected, but in fact is an inevitable consequence of the presentation of the sharpness data in terms of line pairs per picture height (and thus independent of format size).

Quite simply, at any given focal length and aperture, the lens will have a fixed MTF50 profile when expressed in terms of line pairs per millimeter. In order to convert to lp/ph, we have to multiply by the sensor height (in mm); as the full-frame sensor is 1.6x larger, MTF50 should therefore be 1.6x higher.

In practice this is an oversimplification; the AP tests measure system MTF rather than purely lens MTF, and at frequencies close to Nyquist the camera's anti-aliasing filter will have a significant effect in attenuating the measured MTF50.

SCHEIMPFLUG PRINCIPLE


Another possible cause of confusion as to how a lens-based OIS negative lens element shifts the image relates to the manner in which tilt & shift lenses work. In a situation where the subject is not square to the image plane of the camera the perspective vanishing points project perpendicularly to the x,y camera tilts. This may be rectified by tilting the film or sensor in such a way that a line drawn orthoganaly to the optical axis through the lens centre, intersects the image plane at the same point as the subject plane. This is known as the Scheimpflug principle. A misimpresion is created that tilting the camera lens tilts the optical axis itself, rather than the image plane, because a tilt & shift lens on a DSLR cannot quite meet the Scheimpflug conditions since only the lens can be tilted not the film or sensor plane.


EXPERIMENT


It seemed the best way to see how the misguided notion that lens-based OIS can compensate for the angled movements of a telephoto lens in a way that a sensor-based OIS cannot, was to put it to the test.

In November 2009 Sigma brought out an autofocus optically stabilized 70-300 telephoto zoom, in Sony alpha fit, the AF 70-300F/4-5.6 DG OS <AF 70-300F/4-5.6 DG OS>. In early March 2010 I purchased one from Park Cameras and conducted an experiment.

I live close to the Royal Lytham Golf Club and have a clear view of their flag staff. I photographed it with the lens set on 300mm @ f/5.6, tripod mounted OIS off, and then handheld, with both lens and body OIS on, lens only, body only and no OIS, and compared the results. The aspect ratio was 3:2 6048 x 4032 pixels. flagstaff map of area
click here for images

Knowing the distance from camera to top of flag pole was 1538.5 feet (obtained by GPS) and by measuring the amount of blurring in pixels per picture height x 1000 (ppphE3), I was able to calculate the image shift of the lens, and the OIS enhancement factor. I obtained radically different results to those presented in the AP test. table of results

With the lens tripod mounted and the shutter fired by IR remote, shutter speeds 1/500s & 1/350s, there was only 3 pixels blurring corresponding to 0.74ppphE3.

With no OIS there was 180 pixels blurring @ 1/60s or 44.64ppphE3.

With the Sigma lens OIS off & Sony sensor OIS on, shutter speed 1/90s blurring averaged @ 3.7ppphE3 a 11 fold enhancement over no OIS, & @ 1/60s 5.1ppphE3 a 7.8 fold enhancement, over no OIS.

With the Sigma lens OIS on & Sony sensor OIS off, shutter speed 1/90s blurring averaged @ 1.98ppphE3, a 21.5 fold enhancement over no OIS. & @ 1/60s 1.1ppphE3 a 39 fold enhancement.

With both OIS systems on (not recommended as they may conflict) @ 1/90s blurring averaged 1.24ppphE3 a 35 fold enhancement. & @ 1/60s 1.74ppphE3 a 24.7 fold enhancement.

The angular movement of the camera tripod mounted was only 12".2arc. The best frame with the Sigma lens OIS on was 16".3arc, and with only the Sony sensor OIS on, 24".4arc. With both OIS systems on, 20".3arc and without OIS 731".6arc.

The angular movements of the lens, when hand held, will be the same with either OIS functioning. The lens-based OIS was most effective by between a factor of ~2 & ~5. With both OIS systems functioning the enhancement compared to sensor-based OIS only, was a factor ~3.

These results do not support the notion that lens-based OIS compensates for angled lens movements in a way sensor-based OIS cannot. What they do quite effectively demonstrate is that lens-based OIS stabilizes the image slightly more effectively than sensor-based OIS. One can speculate why; perhaps it relates to system inertia, whereby the lower mass lens-based unit can shift the lens more rapidly than the more massive sensor-based unit.

What was striking about Sigma's image stabilization was the way the viewfinder image was also stabilized. I was using the Sony L focussing screen which has frame and AF grads, and a centre AF box. These light up red when the AF kicks in. I could hand hold the zoom lens @ 300mm and keep the flag pole button (the cap placed on the top of a flag pole) in the central box. With no OIS I found it impossible to do so, the button moved around the central box and I couldn't hold it within the box for even a fraction of a second.

The results I obtained clearly demonstrate that with either lens-based or sensor-based OIS you could increase the hand held shutter speed by 2 stops. I suppose in bright light conditions you could extend it by between 3 & 4 stops as the manufacturers claim. But based on the reciprocal focal length rule for the minimum shutter speed (i.e. 1/300s) I could not obtain acceptable images @ 1/30s (3 stops) or 1/15s (4 stops). They were just as blurred as the images obtained with no OIS. I intentionally chose a scene that was dull. Heavy grey overcast sky filled with low stratus, at sunset. It was just gauged to allow use of 1/300s @ f/5.6 for the tripod mounted images, which I shot first. A little later as the sky darkened, the exposure times had to be increased to 1/90s & 1/60s with the lens wide open.

It also struck me that when you own a Sony a900 and use Sony fit image stabilized telephoto zoom lenses you have the best of both worlds. You have the lens-based OIS when you need it, and for lenses without image stabilization, you can use the sensor-based OIS. Either way you cannot loose.

HANDHELD SHOOTING TECHNIQUE

Pointing a telephoto lens and trying to hold it on a specific small distant object is not easy. Keeping the object centred becomes easier with practice. If you take up rifle shooting or bird watching you eventually become an adept. Years of using hand held terrestrial refracting telescopes has made me adept. Not only can you point a long tube refractor readily on target and hold it there, you can do so for much longer. It was evident when I was demonstrating my replica of Galileo's "Old Discoverer" to members of our local astronomical society, that those with little observing experience, found it almost impossible to acquire the target, let alone hold it.

The Sigma AF70-300F/4-5.6 DG OS zoom is not particularly heavy or lengthy. I was curious to see how long an exposure I could get away with and still get an acceptable image. I photographed the dial of a Banjo Barometer and examined its Compass Rose. All the images were shot at 300mm f/5.6, from 1/15s thru' 1s, varying the ISO rating from ISO6400 to ISO400.

The results were very interesting. Both lens-based and sensor-based OIS gave an acceptable sharp image at 1/15s. The lens OIS gave a slightly sharper image @ 1/8s. Both lens and sensor OIS gave an acceptable image @ 1/4s, & lens-based OIS won out @ 1/2s, whereas neither could cope with a 1s handheld shot. click here for images

The Sigma lens OIS and Sony sensor OIS could compensate for up to 1/4s exposures, the lens-based OIS winning out by just 1 stop @ 1/2s. What this little experiment demonstrated is that sensor-based OIS is almost as effective at compensating the angled movements of a compact telephoto. When shooting in bright sunlight I don't think you would be able to tell the lens-based or sensor-based OIS results apart. The big difference in hand holding the lens is the viewfinder image stabilization. I take back what I said in the opening section of this article, it may not provide a marked framing advantage but it certainly makes holding the frame a lot easier.

POSTSCRIPT

I have been a regular reader of Amateur Photographer since 1964. This magazine has a fine pedigree. It recently celebrated its 125th anniversary. Every test report I have read since 1964 has been technically excellent. This particular test report which was admittedly challenging, would benefit from a follow up test using Fx format DSLR's.



Chris Lord

This page was created by SimpleText2Html 1.0.2 on 05-Mar-2010

Top of Page
Return to Home Page