Leica M9, part 2, performance compared to Nikon D3x.
Introduction
The M9 landed in the Leica sphere with a bang. As is usual with splashes, they calm down quite soon. And then we have time to reflect on the long term impact and significance of the M9. Historically the introduction of the M3 can be seen as offering the closest parallel. In those days we did not have the internet rumor machines and blogosphere. The birth and development of the M3 was only known to a small group of persons. The only information channels were the printed press and the photo magazines. The pace of innovation was much more relaxed and one could take the time to think and reflect on important issues and themes. The Leica M3 offered un unprecedented ease of use, the highest level of precision engineering in photography and the best lenses. This package did boost the quality and immediacy of documentary photography and for a brief period of time the M3 was the most valued camera model in the world.
Switching from a fifty-year-old M3 to a two-weeks old M9 does not force upon the user a paradigm shift in handling and concept. In fact the newest branch on the evolutionary Leica tree is an M7 with a digital core. And the M7 is the M6 with a core of automation and the M6 is an M4 with an inbuilt-exposure meter. And the M4 is an M3 with upgraded finder. This sketch of the Leica M lineage is anathema for Leica historians who attach prime importance to small engineering differences, but for the attempt to locate the M9 in the Leica world, it will suffice.
The Leica M9 will presumably not have a comparable impact as the M3 had. The world of photography is too different now. And the world of engineering and manufacturing has changed too. Many camera historians will claim that a Zeiss Contarex, a Leica M3, a Nikon F and a Canon F1 represent the pinnacle of mechanical engineering sophistication. Compared to, for example some current MicroFourThirds models, the cameras mentioned above, appear to be rather crude in design and construction. The Leica M9 stands firmly in the classical precision engineering philosophy. That is the strength, but also a weakness especially when we want to look at the future developments. I will return to this theme at the end of the M9 report, but for now I will look at the capabilities of the M9 as they are.
The IR issue
The coupling of a solid state sensor in the M camera body with the Leica lenses with the typical short back focal distance of 27.80 mm posed big problems for Leica designers. Many digital camera systems employed several layers of filters in front of the sensitive pixel surface to cure problems of moire and resolution, resulting in a filter thickness of up to 4 mm. This is a hefty thickness and will certainly have an effect on image quality. The steep angle of incidence of the rays at the edges of the frame will interfere with the thickness of the filters and result in increased shading and color shifts and reduced definition. Given the state of the art, Leica decided to focus on a layer as thin as possible to improve image quality. The consequence of this decision was a reduced sensor size and an increased IR sensitivity of the M8 series of cameras. An additional IR filter on the lens effectively solved the IR problem, but is not the most elegant solution imageable.
I used a test chart specifically designed to show IR color casts and tested the M8 with and without a filter, the M9 and as comparison the Nikon D3x, the current camera of reference in the high end camera market.
Below you find the results for M8 with and without filter, the M9 and the D3x. Note that the M8 with filter solved the problem completely and the M9 is totally free of IR artifacts, as is the D3x. The M9 filter thickness is now 0.8mm as compared to 0.5mm for the M8 and more than 2mm for the D3x. The Nikon cover glass incorporates a lowpass filter that the Leica lacks. Leica claims that the solution in the M9 improves definition of fine detail. The occasional occurrence of Moire effects are filtered by the software. This applies only to the JPG images, which you often do not want to use because of the internal processing that is not always as good or flexible as the best RAW developers support. Here we find the classical difference between lab processing and your own darkroom processing.
Below you find four pictures in a small format, but the relative sizes are identical to the full filesize.
Note the comparatively petite format of the M8 images. Original sizes have been scaled to 70%.
First the M8 image without filter.
Then the M8 with filter. Note the very different reproduction of the black materials, specifically selected for their IR reflectance. The IR flter is commendably effective end M8 users cannot complain.
Next is the M9.
You cannot really be negative about the M9 images. There is not a trace of recorded IR reflectance. The filter before the swnsor is quite effective and there is also no loss of definition.
Below is the Nikon image.
Note the clearly visible larger area which is important Again we may refer to the classical wisdom of AgX times: a bg format with a moderate lens may give better results than a small format with an excellent lens.
The definition of the sensor.
The M9 sensor has pixels with a size of 6.8 micron and the Nikon D3x has a pixel size 0f 5.9 micron. Theoretically the Nikon has a small edge in ultimate resolution, but we should again return to old wisdom and note that there is no substitute for size (a bigger capture area or negative is always a definitive advantage) and a better lens can offset a bigger capture area to a certain extent. A CRF camera has additional aspects to look at to for optimum performance. The camera lens must have the exact distance from lens flange to sensor location, the rangefinder must be adjusted to this distance too and the focus cam must be precisely machined with the correct steepness over the whole focusing range. Leica sets the nominal distance between flange and sensor location to 27.80 mm. I checked three lenses (3.8/24, 1.4/35 and 1.4/50) and got these values: 27.84, 27.83 and 27.81! This is an outstandingly good result and a fine indication of the care of production at the Leica factory. There is an age old rule that states that with smaller apertures the accuracy of focusing can be relaxed as the extended depth of fileld will compensate for focusing errors. I will explain in a parallel article that the introduction of the M9 forces one to rethink the classical testing paradigm, but here I will look at two aspects. I did a rather exhaustive analysis of the focusing accuracy of the M9 with the 1.4/35 asph (assuming that this lens will take a prominent place on the M9 because of its classical focal length). I placed the M9 in front of the familiar test chart (the nine star charts in three rows of three columns) at a distance of 1.35meter. I again assume that the Leica user will exploit the format size to its optimum and will get as close to the scene as possible. At closer distances the focusing accuracy is more critical. I made a range of eleven exposures with through focus steps of 1 cm (from 1.30 to 1.40) on the tripod. One range at full aperture and one range at aperture f/2.8, where one would assume to have some latitude in focusing errors. Then I evaluated every star in the center with the Imatest analysis program and selected the best result. With wide open aperture the result is amazing: the zero position delivers the best result. Note that both extremes are less good The plus range gives generally the better performance: so it is best to focus a bit farther than indicated. With the aperture stopped down to f/2.8 the best position is the +2 focus: here we see in critical situations the effect of a slight focus shift. BUt it is also evident that even at 2.8 there is hardly room for focusing errors if the optimum performance is required The accuracy of the rangefinder of the M9 is beyond reproach, but the optional magnifier (1.4x) is absolutely necessary. Find below the full range of graphs for the eleven steps with the lens wide open at f/1.4.
-5
-4
-3
-2
-1
0
+1
+2
+3
+4
+5
This exercise was repeated for the Nikon D3x (not shown here) and the best focus position on both cameras at 1.4 and 2.8 was registered. Then the same test-chart were shot and the result now evaluated for maximum definition and resolution.
See below the comparison first at 1.4 and then 2.8.
To study the effect of sharpening in post processing I used Capture One with sharpening at zero and 100%.
Note that the result of the sharpening process is to enhance the contrast of the low and medium frequencies and has hardly any effect on the critical high frequences.
Leica at 1.4 without sharpening
Leica at 1.4 with sharpening
Nikon at 1.4 without sharpening
Nikon at 1.4 with sharpening
Leica at 2.8 without sharpening
Leica at 2.8 with sharpening
Nikon at 2.8 without sharpening
Nikon at 2.8 with sharpening
We may infer from these graphs that the Nikon images benefit more from the post processing, but get that now familiar (and not always pleaseant) digital look, where the Leica images are more closely related to the classical film look and here the sharpening effect is less pronounced. The Leica M9 is quite close to The Nikon D3x in definition and resolution, but Nikon photographers do not need to fear that the M9 will dethrone the D3x as the reference camera for state of the art quality. Stunning as the M9 pictures are, they must be put in context and then the Nikon D3x images are just better.
Read part 3 here
Added: how to read the graphs.
The tests are done by photographing the Siemens star pattern. This is a pattern of alternating black and white pie-shaped segments. From edge to center the the lines get smaller and smaller and then wil merge into a grey area. The transition from white to black is not abrupt but follows a sinusoidal pattern, suitable for MTF analysis.
See below an example of this chart.
The program divides the chart in eight parts in order to analyse the several line orientations from vertical to horizontal.
In the charts this is indicated as segment 0 to segment 8. For every segment the contrast transfer is calculated and presented
in the classical way as a graph. The graph runs horizontally from left to right (zero linepairs to maximum linepairs) and vertically from bottom to top (zero contrast to maximum contrast). The program used does not read the linepairs/mm but the linepairs per picture height. Sometimes the values used are cycles/pixel or linepairs per pixel size. But that is not important fro the analysis here. The maximum frequency is limited to the Nyquist frequency and that is indicated by the vertical blue line with the 'Nyq' tag. You can see that the Nikon image has a higher maxiumum frequency than the Leica image.
In the ideal world the graph would be a staight horizontal line at the top of the chart, representing maximum contrast over the full frequency range from subject outlines to fine detail. The chart can be divided in three broad bands: left covers the subject outlines, in the middle you find the finer detail and at the right you find the textural details. Beyond the Nyquist line, the curves meander wildly: that is the effect of the spurious resolution and should be neglected.
In the practical world you will see a gradual drop of the contrast when you move from low frequencies to high frequencies. The steepness of the graph and its shape indicate the performance of the lens. Normally a lens will show a different performance in the sgittal and tangential direction: you see this when the several graphs for the segments are diverging. If the lens is well corrected the graphs lay on top of each other and form one line.
Generally you can say that the best lens has the largest area under the graph or is located as far as possible to the right of the graph. If at 1000 lp/ph the mtf is 60% for one lens and 50% for the other, then the first lens is better in this respect.
There are two sets of graphs: one where no postprocessing sharpness has been applied: here you see the gradual drop in the shape of the lines. The other set of graphs shows the result when postprocessing has been applied and then you see the characteristic bulge in the shape at the lower frequencies. Most postprocessing sharpness adds artificial contrast edges to the outlines of the subject, whch gives a crisp, even brittle view of the scene. This looks good to the eye, but it is just perception. In the graphs you see this effect clearly.
For comparision:
here is the graph of a low contrast star pattern. Note the steep fall off.
here we have a high performance graph
And here one with sharpening applied. Note the boost in the mid frequencies, but the same pattern for the fine detail representation.
