Stephen Burch's Birding & Dragonfly Website
The EOS 50D fared badly in the noise comparisons mainly due to its smaller pixels, so one aim of these resolution tests was to investigate if there was an upside to these small pixels - did they lead to more detail being visible in the images? In other words, is there some benefit to 15 Mp, compared with the 10Mp and 8Mp of the 40D and 350D, to counterbalance its poor noise performance?
On the other cameras, I again made sure any similar options were off, although there are fewer available on the older models. Having done this, all the raw images should be as near as possible to what is coming off the sensor, with minimal processing. In other words a like for like comparison between the cameras, or 'level playing field'.
For these resolution measurements I prepared and printed out a resolution test chart, illustrated below. The full version can also be downloaded from Norman Koren's excellent website, which also gives full instructions on how to use it to derive MTF curves.
Note that I used the 5mm version of his test charts, which are intended for DSLR cameras, and when printed at 25cm width, have a magnification of x50. However with my dated inkjet printer, the pattern 'died' for anything to the right of about 100 lp/mm, and one I had commercially printed was little better. I understand the best current inkjet printers can do better - and are sharp across the full chart.
Norman Koren describes a procedure for MTF evaluation, which requires the test chart to fill a certain fraction of the field of view, which will of course be different for different lenses. But when used at this distance, I found the camera/lens combinations out resolved the chart!
I was however less interested in MTF information in terms of lp/mm on the camera sensor, and more interested in comparing the resolving power of different lenses and cameras. So I found it best to move the chart further away. After a few experiments, I found a distance of about 25m was suitable for my purposes. At this distance, even the highest resolution camera/lens combination did not out resolve the printed pattern. All the tests were made at this fixed camera to chart distance - finding somewhere suitable to get 25m from your chart is not that easy!
Incidentally I find it a remarkable testament to the quality of these cameras and lenses that the target had to be so far away.
Having found a suitable location, it was then a matter of making the measurements I was interested in. To minimise other effects such as camera shake, I did the following:
With all the above, the results were reasonably reproducible, and the variations within the shots for the same camera/lens combinations were small.
Back on the computer, I then converted all the raw images into tiff using DPP, and then magnified the images up to 100% - 200%. For each one, I simply estimated the limiting resolution achieved (i.e. the lp/mm value at which the black/white pattern could no longer be distinguished).
A typical image is shown below:
For each lens/camera combination, I repeated the above about 10 times, and then took the average limited resolutions measured.
*Chart width on sensor, for the 400mm lenses without the x1.4TC, was 3.8mm, not the 5mm required for absolute MTF values.
In making these measurements, I estimated by eye the point at which the contrast due to the lines on the chart fell to zero, i.e. the point at which no clearly distinct lines could be discerned. This was a somewhat subjective judgement, and others may derive different results from the same information. On the whole, I think my values were 'conservative'.
The limiting resolution values given in the table above are specific to the magnification I used when taking the photos of the resolution chart. To derive absolute MTF values, the full width of the chart should have been 5mm on the camera's sensor, whereas for my shots without the x1.4TC, its width was only 3.8mm. The table above also gives the limiting MTF figures, derived by multiplying the limiting resolution values by 5/3.8=1.32. To obtain MTF values for the x1.4TC, it was also necessary to divide by the extra magnification provided by the TC (i.e. 1.4).
The column headed Nyquist frequency is the limit on the maximum achievable resolution set by the number of pixels in the sensor. My limiting MTF values are mostly about 80% of the Nyquist frequency. Some of the photographs show evidence of beating and aliasing effects at higher frequencies, which suggests the 400mm lenses assessed here are not the limiting factor - the limits are instead caused by the number of pixels.
To obtain the actual lp/mm on the chart, the above values would need to be divided by the magnification of the chart (x50.4 in this case). So for the highest measured resolution of 84 lp/mm on the chart, this corresponds to a remarkable 1.7 lp/mm at a distance of 26m!
From the above table three things are clear immediately:
Another way of presenting these results is to plot a graph, which is often easier to interpret than a table of numbers.
It seemed that the most meaningful parameter to examine was something related to the expected resolving power of the lens/camera combination. Resolving power should be proportional to both the focal length of the lens and the number of pixels across the image. Note that when using the x1.4 TC, I used an effective focal length of 1.4 x 400 mm = 560mm. A dimensionless "resolution" parameter that also takes the sensor size into account is (lens focal length)/(pixel pitch). For the 50D sensor, the sensor width is 22.3mm, and there are 4752 pixels across an image, so the pixel pitch = 22.3/4752 mm = 4.7 micron.
The graph below shows the measured resolution values, relative to those to the 350D 400mm f5.6 value, as a function this dimensionless resolution parameter. The straight-line through the origin is a best fit to the values for all lens/camera combinations apart from the point furthest to the right (50D, 400f4DO & x1.4 TC).
This graph shows that all but the point furthest to the right fit on a straight-line, i.e. the resolution achieved is very close to that expected on the basis of the focal length and pixel pitch.
So the smaller pixels on the 50D are giving the maximum possible benefit, at least for the 400mm f5.6 and 400mm f4 DO lenses. The one case which shows less than expected benefit is adding the x1.4 TC to the 50D and 400mmf4 DO lens. In this case, the TC increases the resolution by only about x1.3, compared with the hoped for x1.4.
This is a slightly disappointing result, but it probably simply shows that the inherent resolution of the 400mm f4DO and x1.4 TC is lower than that achievable with the 50D sensor. However, my 50D when used with the 400mmf4DO and x1.4 TC requires a very large microAF adjustment, and there is the possibility that something connected with this strange problem might be limiting the resolution achieved here.
Another possibility is that the printed test chart is also begining to 'die' around this resolution, and that if I repeated everything further away, the improvement factor might appear to improved. But looking carefully at the test chart, I think this is unlikely - the pattern is quite clear to beyond 100 lp/mm.
I haven't yet tried the 400mm f5.6 with the x1.4TC, because I no longer use this combination, and it needs the x1.4 TC pins taped to give any sort of autofocus. It could be something worth trying in future to see how it compares with the 400mmf4DO.
It is worth adding that the graph above should also be applicable to any DSLR camera/lens combination including full frame cameras such as the EOS 5D, provided of course the optics match the sensor's capabilities.