Leica M9, part 3, definition M9 versus M8 and D3x.
Introduction
There is a valid question to ask about the Leica M9 tests in general. The final photograph (on print preferably or on the screen) is the result of a chain of processing steps. Specifically you cannot test a sensor in a camera without a lens attached to the body. In fact you can never make statements about the sensor quality or performance without reference to the lens attached. This is not a new observation: in the days of AgX emulsions film was tested with a lens on the camera. And everyone was aware or could be aware of the fact that lens and film had a cascading relationship: when the lens had a quality factor of 0.7 and the film had a QF of 0.9 the resulting image could have a maximum of 0.6 of the theoretical optimum. Film emulsions could be tested without a lens (with a grid or a knife edge directly placed on the emulsion) and one could get resolution or MTF figures in a stand-alone fashion. For lenses the same approach worked: you can measure the MTF or other properties of a lens without using film.
In the current digital world this approach does not work anymore. The interaction between sensor and lens is much more complex and multi-facetted than can be inferred form the rather crude but ubiquitous LP/PH figures. It is too simple to look at the sensor data and find that there are, let us say, 4000 (wide) by 2000 (high) pixels on the sensor matrix and then to divide this number by 2 and infer that the maximum resolution is 1000 linepairs for this picture (sensor height). This number is also referred to as the Nyquist limit or limiting frequency. You need a minimum of two pixels to record two lines with different luminance to see a line edge. To be more precise we have to say that sampling in the spatial domain only refers to edges: an edge in the spatial domain is the equivalent of the sinusoidal shape in the temporal (audio) domain. Just counting LP/PH is the same approach as looking at resolution figures that was so popular in the sixties and seventies of the previous century.
The Nyquist rule is properly identified as the Shannon or Nyquist sampling theorem. The sampling theorem indicates that a continuous signal can be properly sampled, only if it does not contain frequency components above one-half of the sampling rate. For instance, a sampling rate of 2,000 samples/second requires the analog signal to be composed of frequencies below 1000 cycles/second. If frequencies above this limit are present in the signal, they will be aliased to frequencies between 0 and 1000 cycles/second, combining with whatever information that was legitimately there.
The Leica M9 sensor has 145 pixel rows per mm or 6.8/6.9 micron per pixel. The number of pixel rows is equivalent to the sampling frequency. The Nyquist theorem tells you that the analog (optical) signal must have frequencies below 73 c/mm to be properly sampled. That is the reason why most camera sensors use a low pass filter to cut off all optical signals above this frequency. A low pass filter does not have a sharp cut off but a gradual suppression of high frequencies. And many lenses can record frequencies above 70 lp/mm. The quality of the edge will be influenced by all these aspects. Not unlike the quality of the grain in an emulsion is more important than the mere number of grains per square mm. Instead of counting number of lines it would be better to study the quality of the line edge!
Schneider and others have demonstrated that the simple rule that two pixels will suffice for the recording of a pair of black and white bars (a line) only applies in the case the thickness of the bars is identical to the pixel pitch and match completely. This condition is not often found in practice and then we may need three or more pixel rows to record an edge.
To follow the question stated at the beginning one would ideally one like to select a lens that has a limiting frequency equal to what is required by the Nyquist sampling theorem. Many Leica lenses can resolve more than 80 lp/mm and these would be aliased to frequencies below 70 c/mm and influence the quality of the edge.
These aspects are unavoidable and more study is needed to look at the detailed reproduction of the edges. For the moment the best one can do is to select a lens with very good definition that is representative of the performance of the Leica lens range. Results that are presented should always be interpreted as system results of the cascading of lens/sensor/in-camera image processing.
System analysis
The systems that are analysed in this report of the M9 (the M9, M8 and D3X) are all high grade systems and it is not easy to make general statements. There is a tendency, sometimes an imperative to make statements about the best system, where ‘best’ is interpreted in maximum number of LP/PH or some other quantifiable parameter. It is evident that the M8 (as example) will not score very high marks in this discipline because of its smaller sized sensor. The practical examples below will show that the M8 will resolve less detail than the other two cameras. It cannot be the case however, that a camera that a few months ago was characterized as producing outstandingly good imagery now is ripe for the dustbin. This attitude would imply that a car that has a top speed of 100 m/h would be useless when a new model arrives on the market with a top speed of 120 m/h.
The current behavior of looking at image files at 100+% on a computer screen and study individual pixels has aptly been called pixel peeking and has hardly anything to do with appreciating or evaluating photographs or pictures. You did not study grain patterns under the microscope in AgX days, did you?
And in the AgX period there were films with different grain size and resolution. Microfilms record a stunning amount of detail, but you can hardly see it unless at unpractically large prints. When using normally sized prints in A4 or A3 format many recorded details cannot be printed and are factually invisible. Pixel peeking at 100% on the screen is the equivalent of grain viewing under the microscope. But is not the same as creating prints for appreciative viewing. We should never forget to ask what we can effectively see in print and what we want to record in print. This is obviously not the same. Some common sense and sanity is certainly needed in the digital domain. The same is true with the M8 and M9 pair. The M9 would represent a microfilm approach in AgX terms and the M8 a 100/400 ISO film.
Below I have selected some representative areas from the test picture used in installment 2 of this series.
The pictures are processed by Capture One with very moderate sharpness parameters and no adjustment of color or exposure. The export was a large TIFF file, selections were made and converted to JPG for web-use. There is some quality-loss compared to the original TIFF files, but these are too big for web-use.
This is the top left corner of the D3X picture.
The quality is very high, with accurate colors, good definition and no artefacts.
This is the same selection from the M9 picture.
The quality is comparable as far as definition goes, but Moire artefacts are clearly visible and color shift is present too.
This is the M8 picture at original scale.
The quality is high with quite good color, moderate definition and no artefacts.
The same picture, but now scaled to 133% to get the same size as the original M9 image shows no improvements of course, but the differences are now more clearly visible. Especially the lower contrast of the M8 image is remarkable.
With the introduction of the M8 Leica claimed that the smaller sensor size was necessary to take account of the short back focal length and the steep angle of the oblique rays. The M9 improvements are impressive, but the comparison with the D3X images shows that even with a low pass filter and a thick system of cover glass in front of the sensor Nikon engineers can deliver first class imagery at least comparable with the M9 performance. The Moire artefacts of the M9 are peculiar while the M8 is free from these artefacts.
Below is a picture from the center of the image. This part of the test image is the Esser test chart for the comparison of difficult and accurate skin tones. But you can also look for sharpness details in the hair.
The D3X picture is below. Sharpness is excellent and color reproduction is very good. There is a slight over exposure.
Then the M9 picture. Sharpness is slightly lower and the skin tones less accurate. But the high lights are better preserved, which could be attributed to a high dynamic range.
Below is the M8 picture. Color reproduction is close to the Nikon example. But overall the quality is good, but it lacks the bite of the other cameras.
The third example shows the limits again. The M9 and D3X are very close, but the Nikon images are somewhat cleaner and a tad crisper. Careful post-processing can minimize the differences. The M8 picture shows a lack of highlight detail, an indication of a narrow dynamic range. This may be caused by the compression algorithm used in the M8.
D3X below
M9 below
M8 below
Overall
The M9 has joined the top class of 35mm sized sensors, but with some caveats. The artefacts may not be visible on all occasions, but show that Leica should do some additional home work in this area. The smaller sensor size is not a drawback when coupled with the excellent Leica lenses, but the absence of the low pass filter does not bring any truly competitive advantages. The M9 pictures are not better than the comparable Nikon pictures, but presumably at this stage of the evolution one could not ask for more. The M9 user does not have to entertain any inferiority feelings, but a realistic assessment of strengths and weaknesses is not a bad thing. In the course of this series we will look at other aspects of the M9.
M8 versus M9.
The comparison in relative and absolute terms between the performance of the M8 and M9 will command much space in print and discussion among Leica aficionados. In the series of pictures discussed in this article the M8 came out on third place which is no surprise, but a bad result it is certainly not. In the tests the full size of the frame was exploited and this means that the M9 can be put closer to the scene with the same angle of view but a bigger enlargement allowing more pixels to be used for the same scene detail area. This is a bit unfair, as if you were comparing a half frame (PEN) camera with a full frame (M) camera (in classical analog terms). This theoretical disadvantage of the small Leica negative was already noticed by the early Leica pioneers when comparing Leica enlargements with 6x9cm contact prints. Their solution was the rule to get as close as possible to the scene so that there was no cropping necessary when enlarging and every precious grain clump could be used for the print. So I set up the M8 in front of the test chart and chose a distance that fills the sensor area with the total area of the test pattern. Then I kept the distance, used the same lens and now the M9 was on the tripod. Because of its bigger sensor, the area of the test pattern does not fill the sensor area but only the center portion. This center portion has the same dimensions as in the M8. The lens is identical, the distance too, so the enlargement factor is the same and the pixel pitch of both sensors is also the same. If you crop the M9 image to the dimensions of the M8, you have two identical pictures in scale and area. These pictures (M9 cropped and M8 full size) show the test patterns on the same scale and with identical pixel dimensions. The pixel pitch of 6.8 microns does not change between the M8 and M9.
Below is the pair of graphs for the M9
center M9
edge M9
The center graph shows a very good result with almost film-like properties. Compared to the D3X we might even state that the M9 tends in its image qualities to a analog representation while the Nikon has a more digital representation. For Leica cross overs this behavior is very pleasant. And non-Leica migrants might see favor in this combination of film like properties and excellent optical performance.
The Nikon is below and shows the typical digital bump in the medium to low frequencies. Note too that the edges show strong astigmatism, but overall the edge performance is excellent, indicating a very firm command by Nikon engineers of the analog-digital conversion.
Below D3X center
Below D3X edge
Below we see the M8 pair
Below M8 center
Below M8 edge
The center graph shows a moderate sharpness enhancement in between the D3X and the M9. The edges are somewhat better than in the M9, an indication that the thicker cover glass has its effects. It is logical too as the thicker glass forces the rays to follow a longer path through the glass and this reduces contrast and shifts color balance to cyan. The M9 user must be very careful in exercising the technical parameters. Below is a graph of the edge of the M9 image now sightly defocused. The drop in quality is visible.
The interesting comparison is below where we have used the cropped M9 frame.
Below M9 center cropped
Below M9 edge cropped
The resolution is now identical to that of the M8, and the shape of the curve is the same as the shape of the M9 non-cropped, just shifted to the left. These figures demonstrate that the M8 can capture the same level of definition as the M9 when both pictures have the same scale and magnification. With other words, if you do not exploit the 1.33 factor of the bigger M9 sensor to the full, then the M8 might be as good as the M9. Of course the M9 has a higher potential for bigger enlargements, but if you restrict yourself to high quality A4 prints, then the M8 is a good alternative to the M9. I have expanded on this topic as I got numerous questions from readers who are concerned about the status and future use of the M8 in the light of the boosted expectations around the M9. As can be seen in this part of the report, the M9 is without doubt a big step forward compared to the M8, at least performance-wise, but that does not imply that the M8 is not a formidable picture machine.
It has to be stressed that the current attitude of pixel peeking is detrimental to the pleasure and appreciation of real photographic prints, where many of the characteristics that are visible on screen, are not detectible in print. A case in point is the occurrence of chromatic aberration (the red or blue bands next to black/white transitions) which is clearly visible on screen, but not visible in prints with lower magnification or in blak/white images.
The classical approach to lens design has always been to focus on those properties that are visually important on the print or in projection. That is why the limit of 40 lp/mm has been selected as relevant for photographic purposes. In my current blog I have argued that this paradigm needs to be adapted to current practices and demands. But one should not overshoot the case. More on this in the next part.