Pentax shake reduction revisited (test)

**Rondec** · 04-15-2010, 04:03 AM

Certainly looks like something for Pentax to work on. I have to say that I have used quite slow shutter speeds with a 300 mm lens -- like 1/8 second to 1/10 second and had quite sharp photos. Sharp to the pixel level? Probably not, but certainly sharp enough to print 8 by 10 and not notice any blur. Of course, those are the shutter speeds where I am really looking for a difference -- at 1/100 second or better, I am comfortable with my hand holding technique.

I guess the issue is that the sensor can't be moved fast enough to compensate for shake when shooting faster shutter speeds? I know that DP Review does feel like Olympus implementation of shake reduction is quite a bit better than Pentax's, so perhaps that is why.

falconeye · 04-15-2010, 04:57 AM

Originally posted by Rondec

I guess the issue is that the sensor can't be moved fast enough to compensate for shake when shooting faster shutter speeds?

The Pentax SR can move astonishingly fast, supporting focal lengths up to 800mm!

No, the problem is latency aka system response time or frequency response. And this is not an inherent limitation of sensor-shift as the excellent results of Olympus E3 show indeed.

labnut (aka P. Smith) proposes that there is a low pass filter in Pentax' SR system effectively turning SR off above some cut-off frequency and below some shutter speed.

The murata design document I link in my blog article proposes a 1000Hz low pass filter indeed. But this should be ok down to 1/200s actually. Another detail in the murata design document is more cumbersome: the sensor report angular velocities 50 times a second (50 Hz response). So, it may be that some Pentax engineer concluded that the sensors aren't fast enough for much faster than 1/50s. However, this conclusion would be wrong. The main contribution of shake even at very fast shutter speeds is from body tremor which is about 4 Hz only.

This assumes of course that the actual shutter release press is extremely smooth and mirror slap is not an issue.

Because if shutter release press and mirror slap are the major cause of shake within the first 10 ms then the sensors with their 20ms response time are too slow to react indeed. Experiments with MLU and remote release with the foot may clarify this

. However, shutter release press and mirror slap being the major cause of shake is not what the 1/f rule of thumb is teaching us!

One more thing ...

Somewhere I remember to have heard that the Pentax SR actually does try to compensate for the shutter release press shake.

If this is (still) true for the K-7, then we have a problem. Because the attempted compensation would have to be for a "typical" release and the "careful" test release could yield worse than average results due to overcompensation. This should be visible as more blur at 1/200s tests with SR on. The ColorFoto 2008 results for 2008 point into that direction. But not anymore in 2010 for the K-7.

falconeye · 04-15-2010, 05:21 AM

One more thing.

All of a sudden, the new Sigma 70-200mm F2.8 EX DG OS HSM coming out this spring looks tempting.
As does the new Bigma OS.

It remains to be seen if image stabilization of the Sigmas is effective at short shutter speeds.

According to the ColorFoto 2010 test, Sigma does have a problem with the new Sigma 70-300mm F4-5.6 DG OS for Nikon: Enabling OS resulted in more blur at 130mm and 1/200s and 1/100s! Therefore, the affordable 70-300 won't help.

On the other hand, DPR believes to have seen (they fail to deliver decent SR tests) that OS has a very positive effect at 200mm and 1/500s and 1/250s for the Sigma 18-200 and Sigma 18-250.

Note that DPR makes no statement about the issue at hand for the K-7 or K-x. Because they didn't test at faster than 1/50s.

Gimbal · 04-15-2010, 12:11 PM

Originally posted by falconeye

The murata design document I link in my blog article proposes a 1000Hz low pass filter indeed. But this should be ok down to 1/200s actually.

I don't understand you here, why would the 1000Hz limit of the gyro sensor have anything to do with maximum shutter time? (Unless you mean the SR system should compensate in real time for shutter movement, but that seems a bit far-fetched.)

The user, I assume, shakes at pretty much the same frequency no matter what shutter time is used, so the gyro sensors should be fine unless you shake at more than 1kHz.

Also the 50Hz response time, what do they mean by that? The sensor obviously has an analog signal out, and they recommend a 1kHz low pass filter. How on earth are the sensor able to produce an out signal with a frequency over 50Hz if that is it's internal updating speed?

Eruditass · 04-15-2010, 12:23 PM

Great post, I hope the word can get out more so this can be improved on!

falconeye · 04-15-2010, 05:30 PM

Originally posted by falconeye

The above chart shows good combinations of shutter speed and focal length (green) and bad ones (red), taking the SR system into account (mathematically modelled). One of the lines is nothing but the 35mm 1/f rule of thumb.

Users at DPR have asked me to make my SR guide more usable. I added a region for blurry images and using a log-log plot, it is easier to use. I labelled axes in a nicer was too.

Here you go. Enjoy

(click to enlarge)

falconeye · 04-15-2010, 05:41 PM

Originally posted by Gimbal

I don't understand you here

I am in the dark as well here. I was just speculating.

The 50Hz thing. I don't know more than you. I guess the murata device kind of integrates stuff before creating an out signal. So, it may not change significantly if poked faster than at 50 Hz.

eva2000 · 04-16-2010, 10:34 PM

Originally posted by falconeye

Users at DPR have asked me to make my SR guide more usable. I added a region for blurry images and using a log-log plot, it is easier to use. I labelled axes in a nicer was too.

Here you go. Enjoy

thanks for the chart, i'm usuallly always down in <1/15 without flash hehe

Finger · 04-17-2010, 07:09 AM

It's interesting to see that the k-x blur bug or whatever is right in the spot where sharpness goes down in the chart.

ghillindy1 · 04-17-2010, 07:48 AM

The data looks proportional throughout the shutter speeds tested, or am I missing something?

**Rondec** · 04-17-2010, 08:25 AM

Originally posted by ghillindy1

The data looks proportional throughout the shutter speeds tested, or am I missing something?

I think what you notice is that it is proportional up to a shutter speed of 1/100 when there is a plateau and then improvement continues once shutter speed achieves the expected sharp range of 1/focal length*1.5. That means that if you are shooting at 300 mm and your shutter speed is 1/150 second you get minimal benefit from SR. However, once your shutter speed gets below 1/100 second, the benefit really starts to kick in.

falconeye · 04-17-2010, 09:41 AM

Originally posted by Rondec

I think what you notice is that it is proportional up to a shutter speed of 1/100 when there is a plateau and then improvement continues once shutter speed achieves the expected sharp range of 1/focal length*1.5. That means that if you are shooting at 300 mm and your shutter speed is 1/150 second you get minimal benefit from SR. However, once your shutter speed gets below 1/100 second, the benefit really starts to kick in.

The chart is mostly a consequence of Peter Smith's precise measurements.

In Peter Smith's aka labnut's own words, the "minimum" around 1/100s is probably due to a low pass override of the SR signal at faster shutter speeds, not some "bug".

The "plateau" is a result of a regularization of my mathematical model: I just forbid the modelled SR to reduce a blur width (at 50mm) below 0.2px or 1µm as I deemed that to be unrealistically optimistic. Peter Smith's data suggest that blur widths could be reduced even more around the middle of the plateau.

Class A · 05-31-2010, 06:16 PM

The chart suggests that the FA50/1.4 is usable in a significantly wider range of shutter speeds than the DA*55/1.4 (in as much it makes sense to attribute significance to the "tack-sharp" vs "sharp" boundary). Falk, do you expect this result to hold in reality or could it be somehow related to the fact that labnut obtained his results using a 50mm lens?

falconeye · 06-01-2010, 02:14 AM

Originally posted by Class A

The chart suggests that the FA50/1.4 is usable in a significantly wider range of shutter speeds than the DA*55/1.4 (in as much it makes sense to attribute significance to the "tack-sharp" vs "sharp" boundary). Falk, do you expect this result to hold in reality or could it be somehow related to the fact that labnut obtained his results using a 50mm lens?

Two replies:

1. The chart should be read as "how large is the distance to the boundary of, e.g., tack-sharp"?. And then, 50 and 55 aren't that different anymore. The flat line is from an arbitrary regularization parameter anyway

2. I know a lot more now and await Pentax' response before posting news about it. The chart will get updated, it isn't correct for exposure times 1/200s and smaller and will become less oddly shaped for larger exposure times.

Class A · 06-01-2010, 04:08 AM

Originally posted by falconeye

1. The chart should be read as "how large is the distance to the boundary of, e.g., tack-sharp"?. And then, 50 and 55 aren't that different anymore.

That's why I wrote "in as much it makes sense to attribute significance to the "tack-sharp" vs "sharp" boundary". Still, Heidi Klum would tell the DA*55 "you are OUT! Auf Wiedersehen" and the FA50/1.4 "you may leave the runway".

I appreciate that the boundary is somewhat arbitrary and fuzzy, though.

Originally posted by falconeye

2. I know a lot more now and await Pentax' response before posting news about it.

I'll watch this space...

04-15-2010, 05:21 AM	#18
falconeye Veteran Member Join Date: Jan 2008 Location: Munich, Alps, Germany Photos: Gallery Posts: 6,871 Original Poster	Possible use of Sigma OS lenses with Pentax One more thing. All of a sudden, the new Sigma 70-200mm F2.8 EX DG OS HSM coming out this spring looks tempting. As does the new Bigma OS. It remains to be seen if image stabilization of the Sigmas is effective at short shutter speeds. According to the ColorFoto 2010 test, Sigma does have a problem with the new Sigma 70-300mm F4-5.6 DG OS for Nikon: Enabling OS resulted in more blur at 130mm and 1/200s and 1/100s! Therefore, the affordable 70-300 won't help. On the other hand, DPR believes to have seen (they fail to deliver decent SR tests) that OS has a very positive effect at 200mm and 1/500s and 1/250s for the Sigma 18-200 and Sigma 18-250. Note that DPR makes no statement about the issue at hand for the K-7 or K-x. Because they didn't test at faster than 1/50s.

04-15-2010, 05:30 PM	#21
falconeye Veteran Member Join Date: Jan 2008 Location: Munich, Alps, Germany Photos: Gallery Posts: 6,871 Original Poster	SR Chart Guide Originally posted by falconeye The above chart shows good combinations of shutter speed and focal length (green) and bad ones (red), taking the SR system into account (mathematically modelled). One of the lines is nothing but the 35mm 1/f rule of thumb. Users at DPR have asked me to make my SR guide more usable. I added a region for blurry images and using a log-log plot, it is easier to use. I labelled axes in a nicer was too. Here you go. Enjoy (click to enlarge)

05-31-2010, 06:16 PM	#28
Class A Pentaxian Join Date: Aug 2008 Location: Wellington, New Zealand Posts: 11,251	The chart suggests that the FA50/1.4 is usable in a significantly wider range of shutter speeds than the DA55/1.4 (in as much it makes sense to attribute significance to the "tack-sharp" vs "sharp" boundary). Falk, do you expect this result to hold in reality or could it be somehow related to the fact that labnut obtained his results using a 50mm lens? Last edited by Class A; 05-31-2010 at 07:25 PM.*

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Pentax K-X Shake Reduction (not noise from rotation related)	Cedromar	Troubleshooting and Beginner Help	3	02-18-2010 03:27 PM
Pentax K-7 Shake reduction problem	raywheeler	Pentax DSLR Discussion	34	12-01-2009 09:30 AM
Shake reduction	julianactive	Pentax SLR Lens Discussion	2	05-06-2009 04:13 PM
Thank you Shake Reduction!	tcdk	Post Your Photos!	4	09-03-2008 09:29 AM
Shake Reduction	rfphill	Pentax DSLR Discussion	3	07-12-2008 03:29 PM

04-15-2010, 04:03 AM	#16
Rondec Loyal Site Supporter Join Date: Mar 2009 Location: Gladys, Virginia Photos: Gallery Posts: 27,650	Certainly looks like something for Pentax to work on. I have to say that I have used quite slow shutter speeds with a 300 mm lens -- like 1/8 second to 1/10 second and had quite sharp photos. Sharp to the pixel level? Probably not, but certainly sharp enough to print 8 by 10 and not notice any blur. Of course, those are the shutter speeds where I am really looking for a difference -- at 1/100 second or better, I am comfortable with my hand holding technique. I guess the issue is that the sensor can't be moved fast enough to compensate for shake when shooting faster shutter speeds? I know that DP Review does feel like Olympus implementation of shake reduction is quite a bit better than Pentax's, so perhaps that is why.

04-15-2010, 04:57 AM	#17
falconeye Veteran Member Join Date: Jan 2008 Location: Munich, Alps, Germany Photos: Gallery Posts: 6,871 Original Poster	Originally posted by Rondec I guess the issue is that the sensor can't be moved fast enough to compensate for shake when shooting faster shutter speeds? The Pentax SR can move astonishingly fast, supporting focal lengths up to 800mm! No, the problem is latency aka system response time or frequency response. And this is not an inherent limitation of sensor-shift as the excellent results of Olympus E3 show indeed. labnut (aka P. Smith) proposes that there is a low pass filter in Pentax' SR system effectively turning SR off above some cut-off frequency and below some shutter speed. The murata design document I link in my blog article proposes a 1000Hz low pass filter indeed. But this should be ok down to 1/200s actually. Another detail in the murata design document is more cumbersome: the sensor report angular velocities 50 times a second (50 Hz response). So, it may be that some Pentax engineer concluded that the sensors aren't fast enough for much faster than 1/50s. However, this conclusion would be wrong. The main contribution of shake even at very fast shutter speeds is from body tremor which is about 4 Hz only. This assumes of course that the actual shutter release press is extremely smooth and mirror slap is not an issue. Because if shutter release press and mirror slap are the major cause of shake within the first 10 ms then the sensors with their 20ms response time are too slow to react indeed. Experiments with MLU and remote release with the foot may clarify this . However, shutter release press and mirror slap being the major cause of shake is not what the 1/f rule of thumb is teaching us! One more thing ... Somewhere I remember to have heard that the Pentax SR actually does try to compensate for the shutter release press shake. If this is (still) true for the K-7, then we have a problem. Because the attempted compensation would have to be for a "typical" release and the "careful" test release could yield worse than average results due to overcompensation. This should be visible as more blur at 1/200s tests with SR on. The ColorFoto 2008 results for 2008 point into that direction. But not anymore in 2010 for the K-7.

04-15-2010, 12:11 PM	#19
Gimbal Pentaxian Join Date: Apr 2007 Location: Sweden Posts: 2,106	Originally posted by falconeye The murata design document I link in my blog article proposes a 1000Hz low pass filter indeed. But this should be ok down to 1/200s actually. I don't understand you here, why would the 1000Hz limit of the gyro sensor have anything to do with maximum shutter time? (Unless you mean the SR system should compensate in real time for shutter movement, but that seems a bit far-fetched.) The user, I assume, shakes at pretty much the same frequency no matter what shutter time is used, so the gyro sensors should be fine unless you shake at more than 1kHz. Also the 50Hz response time, what do they mean by that? The sensor obviously has an analog signal out, and they recommend a 1kHz low pass filter. How on earth are the sensor able to produce an out signal with a frequency over 50Hz if that is it's internal updating speed?

04-15-2010, 12:23 PM	#20
Eruditass Veteran Member Join Date: Jul 2009 Photos: Gallery Posts: 2,207	Great post, I hope the word can get out more so this can be improved on!

04-15-2010, 05:41 PM	#22
falconeye Veteran Member Join Date: Jan 2008 Location: Munich, Alps, Germany Photos: Gallery Posts: 6,871 Original Poster	Originally posted by Gimbal I don't understand you here I am in the dark as well here. I was just speculating. The 50Hz thing. I don't know more than you. I guess the murata device kind of integrates stuff before creating an out signal. So, it may not change significantly if poked faster than at 50 Hz.

04-16-2010, 10:34 PM	#23
eva2000 Senior Member Join Date: Dec 2008 Location: Brisbane, Australia Photos: Gallery Posts: 239	Originally posted by falconeye Users at DPR have asked me to make my SR guide more usable. I added a region for blurry images and using a log-log plot, it is easier to use. I labelled axes in a nicer was too. Here you go. Enjoy thanks for the chart, i'm usuallly always down in <1/15 without flash hehe

04-17-2010, 07:09 AM	#24
Finger Senior Member Join Date: Mar 2008 Location: Québec Posts: 121	It's interesting to see that the k-x blur bug or whatever is right in the spot where sharpness goes down in the chart.

04-17-2010, 07:48 AM	#25
ghillindy1 Junior Member Join Date: Aug 2009 Location: Plainfield Indiana Photos: Albums Posts: 45	The data looks proportional throughout the shutter speeds tested, or am I missing something?

04-17-2010, 08:25 AM	#26
Rondec Loyal Site Supporter Join Date: Mar 2009 Location: Gladys, Virginia Photos: Gallery Posts: 27,650	Originally posted by ghillindy1 The data looks proportional throughout the shutter speeds tested, or am I missing something? I think what you notice is that it is proportional up to a shutter speed of 1/100 when there is a plateau and then improvement continues once shutter speed achieves the expected sharp range of 1/focal length*1.5. That means that if you are shooting at 300 mm and your shutter speed is 1/150 second you get minimal benefit from SR. However, once your shutter speed gets below 1/100 second, the benefit really starts to kick in.

04-17-2010, 09:41 AM	#27
falconeye Veteran Member Join Date: Jan 2008 Location: Munich, Alps, Germany Photos: Gallery Posts: 6,871 Original Poster	Originally posted by Rondec I think what you notice is that it is proportional up to a shutter speed of 1/100 when there is a plateau and then improvement continues once shutter speed achieves the expected sharp range of 1/focal length*1.5. That means that if you are shooting at 300 mm and your shutter speed is 1/150 second you get minimal benefit from SR. However, once your shutter speed gets below 1/100 second, the benefit really starts to kick in. The chart is mostly a consequence of Peter Smith's precise measurements. In Peter Smith's aka labnut's own words, the "minimum" around 1/100s is probably due to a low pass override of the SR signal at faster shutter speeds, not some "bug". The "plateau" is a result of a regularization of my mathematical model: I just forbid the modelled SR to reduce a blur width (at 50mm) below 0.2px or 1µm as I deemed that to be unrealistically optimistic. Peter Smith's data suggest that blur widths could be reduced even more around the middle of the plateau.

Thread Tools	Search this Thread
	Search this Thread: Advanced Search

06-01-2010, 02:14 AM	#29
falconeye Veteran Member Join Date: Jan 2008 Location: Munich, Alps, Germany Photos: Gallery Posts: 6,871 Original Poster	Originally posted by Class A The chart suggests that the FA50/1.4 is usable in a significantly wider range of shutter speeds than the DA*55/1.4 (in as much it makes sense to attribute significance to the "tack-sharp" vs "sharp" boundary). Falk, do you expect this result to hold in reality or could it be somehow related to the fact that labnut obtained his results using a 50mm lens? Two replies: 1. The chart should be read as "how large is the distance to the boundary of, e.g., tack-sharp"?. And then, 50 and 55 aren't that different anymore. The flat line is from an arbitrary regularization parameter anyway 2. I know a lot more now and await Pentax' response before posting news about it. The chart will get updated, it isn't correct for exposure times 1/200s and smaller and will become less oddly shaped for larger exposure times.

06-01-2010, 04:08 AM	#30
Class A Pentaxian Join Date: Aug 2008 Location: Wellington, New Zealand Posts: 11,251	Originally posted by falconeye 1. The chart should be read as "how large is the distance to the boundary of, e.g., tack-sharp"?. And then, 50 and 55 aren't that different anymore. That's why I wrote "in as much it makes sense to attribute significance to the "tack-sharp" vs "sharp" boundary". Still, Heidi Klum would tell the DA*55 "you are OUT! Auf Wiedersehen" and the FA50/1.4 "you may leave the runway". I appreciate that the boundary is somewhat arbitrary and fuzzy, though. Originally posted by falconeye 2. I know a lot more now and await Pentax' response before posting news about it. I'll watch this space...