Think Again: Stopping Camera Shake and the 1/(Effective Focal Length) Formula

This is the article posted on the website. Maybe we can get a debate started on this. It has been a long time since there has been a good equipment debate on this site.

http://www.the-digital-picture.com/N...spx?News=17252

Sean, I have to say that I think that the pixel density theory is not an accurate assessment and I will explain.

I believe it is the FOV and the relation of the file size that makes camera shake more noticeable. Sure additional pixels make larger files, but it is the size of the picture that makes it noticeable. If you have 1/32 of movement that you see on a 5D III file it might be 1.5 to 2x that on the 5Ds R depending on the direction of the movement.

The increased pixel density of the 5Ds R makes the base file and picture larger. You know the group on here are pixel peepers.
When you zoom in to a 100 percent on a 5Ds R you are looking at it much closer than you would be on an equally framed 5D III or 5D II. Because of this your hand shake will be magnified as well. If you upszie the 5D III or 5D II file to the 5Ds R you would see an increase but the picture would pixelate. The 5Ds R magnify's your camera shake with the larger file.

So how can you get around the magnified camera shake. You can re size your picture smaller and it will not be as obvious.

While there may be a relation to pixel size and the way it shows camera shake, in actual field use I do not see that playing out.
The reason I believe this is the 5Ds R files are never worse than I see with other bodies I have used.
I have had the 5Ds R since August and have about 10,000 frames on it. What I am seeing in actual field use is that I am able to shoot at the same shutter speeds as my old 5D II. While I get camera shake that takes away from the file size they are never worse than the 5D II, and the additional resolution is there to take advantage of.

We see people in other forums saying that you have to have a faster shutter speed or tripod to use the high density 5Ds R. The truth I see is that you have to do those things with all of the DSLR bodies to take full advantage of their potential. The 5Ds R is not different in that respect if you want to maximize your equipment, but I can see the the resolution advantage in my pictures with just normal technique and as I said it is never worse than the 5D III.

If its true that you loose a stop due to shutter speed with the 5Ds R this camera is truly gimped because of its poor high ISO performance.
I am not sure why anyone would buy it. But I can tell you the statement I just made is totally false. If it were true I would trade it off in a second.

Quote:

---

Originally Posted by HDNitehawk

So how can you get around the magnified camera shake. You can re size your picture smaller and it will not be as obvious.

---

Can you also get around it by increasing shutter speed using the suggested formula?

I do not own the 5Ds (yet) but I have been using the 7DII with similar pixel density and I have not noticed issues with camera shake. I am not saying I have perfect technique but perhaps a certain level of experience and technique is another way to address this potential problem.

People were already adjusting the formula for APS-C bodies, saying it needed a * 1.6... or even a * 2 due to the high number of pixels compared to early DLSRs or what was acceptable in film. So yeah, if you need sharpness at the pixel level for a large print, or due to a need for an extreme crop, or whatever, then yes, you need a higher shutter speed.

But as said above, you don't always need that extra detail to be perfectly sharp. You can usually downsize. It's a case by case thing.

And yes, technically it's not because the sensor pixels are small, but because of how little of the image each pixel covers. For the sake of not confusing people though, it's easier to just talk about high megapixel counts or pixel density.

The answer would be yes and that's true of all bodies.

It comes down to the ability to tap the extra resolution. I own the 7D II and 5Ds R now. The pictures are never worse than the lower mp bodies. If you want to take full advantage though it takes stricter measures.

People are misslead when they look at the file that's twice as large.

Quote:

---

Originally Posted by DavidEccleston

And yes, technically it's not because the sensor pixels are small, but because of how little of the image each pixel covers. For the sake of not confusing people though, it's easier to just talk about high megapixel counts or pixel density.

---

That is the part of the theory that doesn't seem to pan out.
If it's true then overall when I am hand holding the high mp body I should be getting worse results than other bodies when comparing equally sized pictures.
In real world shooting I am not seeing that.

Quote:

---

Originally Posted by HDNitehawk

That is the part of the theory that doesn't seem to pan out.
If it's true then overall when I am hand holding the high mp body I should be getting worse results than other bodies when comparing equally sized pictures.
In real world shooting I am not seeing that.

---

If you shoot the exact same scene with both the 5Ds and 5D Mark III bodies, with the same exact movement happening with each shutter click, then the movement recorded will cross more individual pixels with the 5Ds compared to the 5D Mark III. In that way, more motion blur is recorded.

However, if you downsize the 5Ds's native resolution to the size of a 5D III, then some of those motion blurred pixels will get thrown away, thereby making the two bodies similar from a sharpness perspective when the picture dimensions are both set to the smaller size. But on that note, what's the point in having a 50MP body if you're just going to downsample every picture to the size of the 5D III's files? And if you instead crop the image instead of resizing/resampling, the result will be more visible motion blur at the same pixel dimensions (with a tighter framing, of course).

In order to obtain the sharpest possible images when using a 5Ds/5Ds R, you'll need faster shutter speeds than what the traditional 1/focal length gives you.

I know it looks as if I published the post without any input from Bryan, but that is far from the case. Bryan and I discussed the post quite a bit before I fleshed it out, and he proofread it before it went live and gave it the thumbs-up.

As far as he's concerned, this isn't new information. He actually touches on all of the major points in his EOS 5Ds review:

Quote:

---

Because there are more pixels in the same amount of sensor space, camera and subject motion causes subject details to cross over pixels at a faster rate, potentially resulting in blur and a loss of pixel-level sharpness. Because of this, you will find that a faster minimum shutter speed is necessary for handholding this camera (and that image stabilization becomes more important). Similarly, fast-moving subjects may require faster shutter speeds to avoid pixel-level motion blur.

This is the 5Ds change with the biggest learning curve. The old 1/(focal length) rule to determine one's shortest shutter speed for handholding (without the aid of image stabilization) no longer works. Many use the 1/(focal length * 1.6) rule to determine APS-C handholdable speeds. This formula uses the 1.6 factor matching the APS-C sensor angle of view difference, but the higher pixel density of the APS-C imaging sensors is the real reason the faster speed has been needed. The same rule or, better yet, 1/(focal length * 2) is a better base estimate for handholding the 5Ds bodies.

---

Quote:

---

Originally Posted by Sean Setters

But on that note, what's the point in having a 50MP body if you're just going to downsample every picture to the size of the 5D III's files? And if you instead crop the image instead of resizing/resampling, the result will be more visible motion blur at the same pixel dimensions (with a tighter framing, of course).

---

The point is having the extra mp available to utilize. I wouldn't talk to other peoples photography but with mine almost every picture is downsized. If I shoot it with the 5D II or the 5Ds R almost all end up at the same size. I crop it to taste and then post it for view on the net. Or I print to a specific size. The 5Ds R gives the option to print a bit larger or crop a bit more. That only happens occasionally.

Quote:

---

Originally Posted by Sean Setters

If you shoot the exact same scene with both the 5Ds and 5D Mark III bodies, with the same exact movement happening with each shutter click, then the movement recorded will cross more individual pixels with the 5Ds compared to the 5D Mark III. In that way, more motion blur is recorded.

---

If this is true then shouldn't we see this in equally framed same size pictures from both bodies? I am not seeing anything out of the ordinary in my work that makes me think it is worse. To the contrary I am seeing improvement at different levels due to the increased resolution. I know the smaller pixel and motion blur theory is nothing new, its been around for some time. But now I have a body with twice the resolution and the theory isn't holding up in real world shooting.

I am not saying that the smaller vs larger pixel theory doesn't hold merit at a certain level. I am saying that in real world shooting it is not making a noticeable difference.

First the statement I believe is correct; To fully utilize the additional resolution of the 5Ds R one will have to use faster shutter speeds or tripod.

Second the statement I believe is incorrect; The 5Ds R requires a faster shutter speed than the 5D III to avoid camera shake.

Quote:

---

Originally Posted by HDNitehawk

First the statement I believe is correct; To fully utilize the additional resolution of the 5Ds R one will have to use faster shutter speeds or tripod.

Second the statement I believe is incorrect; The 5Ds R requires a faster shutter speed than the 5D III to avoid camera shake.

---

The second statement is, of course, technically false - camera shake is affected by many things, like muscle strength, coffee consumption, earthquakes, etc., but not by the sensor. However, it's absolutely true that the narrower viewing angle of smaller pixels will result in a more apparent effect of that weakness/caffeine/tectonically-induced shake. That assumes you're viewing the images at 100% on the same monitor. Whether or not the difference is apparent in real world shooting depends on the speed of the motion relative to the pixel sizes - if you're using a fast enough shutter speed to overcome motion with smaller pixels, it's also enough to overcome motion with the larger pixels, and you won't see any difference.

Quote:

---

Many use the 1/(focal length * 1.6) rule to determine APS-C handholdable speeds.

---

Interestingly, Canon themselves are among those 'many'. Based on the minimum shutter speeds selected by APS-C vs. FF cameras in Av mode (settings restrictions notwithstanding), Canon has the 1/FL rule for FF and the 1/(FL x 1.6) rule for APS-C baked into their firmware.

I never really bought into the theories of more pixels = more sensitivity to camera shake. That is:

• If all things are kept equal (viewing size and IQ acceptability criteria), then there should be no real difference (HDNitehawks's point I believe). The geometry says the two situations are the same. If you had a column of light hitting the center of a pixel and the shake was small enough that the column of light stayed on the larger low density pixel boundaries, but the same shake splashed over onto the adjacent pixels on the smaller high density pixel; then maybe you can show a difference. But in the real world, the whole pixel is covered with light and any movement will cause light to move between pixels no mater the size.
• Even if you make a bigger picture with the higher density sensor (which is one of its main purposes), then it still does not matter because the picture being much bigger is more likely to be viewed from a greater distance in real world view. Agreed, there are times and reasons for getting right on top of it for inspection, but in that case the shake may matter... however, the cases for that type of viewing is probably not the norm.
• The real problem comes from pixel peeping on computers or heavy cropping.

Now for the caveat to the above - One of the reasons I would want the higher density pixels is to get a sharper looking image at a reasonable sized picture. This would mean raising the bar for IQ criteria. The above was if you consider all things to be equal... including IQ criteria. If you want to have a supper sharp image, or you want to maintain a certain quality level while cropping out a portion of the picture in order to get magnification, then shake is going to matter some. But you are still better off or as good as the low density pixel situation.

I think one of the great practical uses that does not get as much press is in macro photography. I would love to see what kind of cropped detail I could pull out of macro shots, but stability would then be very important. However, now we are talking about the technology allowing for greater scrutiny of IQ... which is a new way of looking at it and defining IQ.

The other aspect is the large stitched shots. In another thread, HDNitehawk started zooming in on someones stitched pictures to look for wild life in a huge stitched picture that covered a huge area. I was really impressed with the resolution. But in practicality, if it was to be printed for viewing, I would standing back to look at it. However, if I had the ability to step up and see really sharp, up close detail, it would be cool. At that point the artistry of the composition goes out the window, and it becomes more of a cool spectacle (so to speak).

Sorry, my age is showing when I use words like "cool".http://community.the-digital-picture...cons/icon6.png

Pat

One way to deal with this is to use a tripod :)

Dave

It would be an interesting experiment if there were a device that could introduce a calibrated amount of camera shake to see how the images would be affected by various degrees of "shake".

Edit: I thought I would summarize my points:

1. Just like 1/focal length vs 1/focal length multiplied by x, the noticeable difference in resolution is likely only observed on large formats, on the pixel level, or when cropping.
2. The linear resolution difference between 5DIII and 5DS(R) is 1.5x
3. The negative of the higher density sensor is really only for movements on the 4.14 um to 6.25 um level on the sensor. After that, I actually wonder if having better resolved blur is actually slightly better.
4. All the above is for ideal conditions. But it has already been shown that the increase of resolution isn't ideal and is heavily dependent on the lens used.

Original post:

Ok...for those pixel peeping, I do it too, but come on...stop it.

I wonder how many pixels have to be blurred together before you have a "noticeable" effect on actual image outputs? I have heard of different thresholds over the years for when the average person can refine detail. This is a function of many factors, but mostly viewing distance. For normal viewing distance I have typically heard that the "average person" can refine 75 pixels per inch (ppi) to 150 ppi. Magazines and Apple ("retinal" displays is build around this concept) have tended for 300 ppi. So that is likely for normal reading distances. Then, some high end printers and some data here have gone to 600 ppi.

So, the 5Ds sensor is 8688 x 5792 pixels while the 5DIII is 5760 x 3840. First of all, resolution is linear, while the number of pixels may be double, the extra resolution is 8688/5760 = 1.5x. So, if applying to the 1/focal length, I would expect it to be 1/focal length x 1.5.

Quickly playing with these different thresholds and the output where the average person would be able to discern differences, for 75, 150, 300 and 600 ppi the 5DsR would be 116, 58, 29, and 14 inches. The 5DIII would be 77, 28, 19, and 10 inches.

I can see where the higher density sensor would be able to better resolve blur/movement just like it can better resolve detail. But that likely will only be noticeable at increasingly large outputs depending on viewing distance.

Looking at this another way. The length/width of a pixel for the 5Ds(r) is 4.14 um while the length/width of a pixel on the 5DIII is 6.25 um. So, for blur/movement that affects:

1. 0 um to 4.14 um, neither sensor can resolve the blur/movement.
2. 4.14 um to 6.25 um the higher density sensor can resolve the blur/movement and it affects 2 pixels or 8.28 um of linear distance on the sensor. While the 6.25 um sensor still cannot resolve the blur/movement.
3. 6.25 um to 8.28 um the lower density sensor (5DIII) is now resolving the blur/movement as it has spilled into a second pixel but now the blur/movement is affecting 12.5 um on the 5DIII sensor but only 8.28 um on the 5DsR sensor.
4. repeat 2 & 3

So the higher density sensor would "see" smaller movements (#2) but lower density sensor would show "more" movement (#3) once the movement becomes big enough. Three pixels of the 5Ds(r) sensor can actually blur together and are actually smaller (12.4 um) than 2 pixels blurred together on the 5DIII (12.5 um).

This would actually be interesting to continue to calculate out, but I have to head to work.

Generally, I think you would either need to be making large prints or crop for this to be important. But I think a 1.5x faster shutter speed should be enough to counteract that initial blurring (step 2). But my experience with motion blur is that it usually is pretty bad, likely impacting many pixels, which I expect to see on both sensors equally as bad.

Of course, this all assumes 100% efficiency in resolving detail as you increase density. Going back to some of Rogers work, the 5Ds had 1.13-1.31x more resolution over the 5DIII and the 5DsR was 1.2-1.47x, which is really pretty impressive considering "ideal" is 1.5x, but also has to come into play when we think about motion blur. In other words, if it is already blurry because of the lens or less efficient pixels, then how much more does motion blur really add?

Last night I read a research paper that did tests on camera shake. Toward the bottom of the paper under "Camera-Motion and Pixel Size Tradeoff" it stated "Therefore, sensors with smaller pixel size will need longer exposure duration and thus introduce higher amount of camera-shake."
scien.stanford.edu/jfsite/Papers/ImageCapture/ICIS06_CameraShake.pdf

To Joel's comment about setting up a test.

I have checked this in actual field use and I was aware of the theory when I bought the 5Ds R.
This is a ritual that I have done several times over the last few years. Sitting in the blind before day break I sit in the blind with my 500mm L and keep checking the light to see when it reaches an acceptable point. Shooting off my monopod and using IS I know from experience that with today's equipment I need 1/250 to get a marginal picture with little camera shake. At 1/500 I can get decent results. When I bought my new 500L II with its improved IS I gained a stop, the speed had been 1/500 and 1/1000. When I bought the 5Ds R the ritual is repeated in the morning and I shoot test shots to find the acceptable levels. With the 5Ds R I lost a stop with ISO, but the shutter speed remains the same.

I have been trying to figure a way to test the pixel theory. Maybe attach some kind of vibrator that has a very low frequency attached to the leg of a tripod? It really needs to be more of a shake.

As Neuro said earthquakes can have an effect. We have had a record number of earthquakes (2177 this year) in our state because of the oil industry. Maybe with a bit of timing I could use earthquakes for the testing.

But from what I am seeing (or should I say not seeing) I believe any effect pixel size will have on camera shake will require far less than 1 stop of light to overcome.