-
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:
- 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.
- The linear resolution difference between 5DIII and 5DS(R) is 1.5x
- 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.
- 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:
- 0 um to 4.14 um, neither sensor can resolve the blur/movement.
- 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.
- 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.
- 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.
