introduction
In my previous report I noted that the Leica M8 feels and operates like a true M-camera, but with enough unique points to support the claim that the camera has its own personality, that fits into the current notions of the digital workflow. Leica has always stated that their main expertise is focused on the realization of high quality imagery with filmbased and sensorbased photographic instruments. In this part I will present an in-depth analysis of the optical properties of the sensor of the M8, the limits of the testing procedures and the often neglected issue of the many factors that influence the feasible results when taking photographs in real-life situations. It is assumed in all reports that the maximum attainable quality is also the quality that can be consistently expected in practice. My testing in very controlled lab-situations indicate that this is a myth. The bandwidth of the quality of the practical results is much wider that one assumes and even the bandwidth of test results is also much wider that the tester will acknowledge.
This second part of my M8 report took a longer time to realize than expected because of a number of themes that I encountered during testing.
Leica has upped the ante with claims that their solution is best when you wish to exploit the inherent optical performance of the Leica lenses.
Therefore a more stringent method of testing was adopted to detect small, but meaningful distinctions. The ubiquitous approach of visual comparison on screen does not suffice, nor can we rely on most types of testcharts. I have reported on these issues in previous posts. One of the (neglected) problems with on-screen visual inspection is the influence of the quality of the screen. This parameter is not often included in the quality equation.
What is the influence of the low-pass filter?
Basically Leica has selected a very thin glass plate before the sensor surface with the argument that a thick glass plate will degrade the optical performance. This is undoubtedly true. A glass plate of 3mm in front of the sensor areas does seriously degrade the quality of the image. We also know that beyond the resolution limits defined by the Nyquist frequency, the fine structures are not recorded with the true (actual) frequencies, but with some false frequency. This phenomenon has been identified as aliasing and moire-effects, actually two different phenomena. The low-pass filter (LPF) has been introduced to cut off the fine frequencies at the limit of the Nyquist frequency in order to reduce the recording of details beyond this limit. There are many variants of the LPF and without knowing more features, one cannot state the actual influence. The basic working of the LPF can be seen below (a matlab simulation). The fine detail is indeed eliminated, but also the contrast has been reduced severely.
The resulting MTF would be quite low and this cannot be accepted. We may assume then that the major companies like Canon, Nikon and Sony, to name a few, will employ all kinds of mathematics to create their dedicated LPF. The question is then not: is there an LPF or not, but what kind of LPF has been employed. In any case the reduction in contrast and the loss of fine detail needs to be counteracted and here the major difference between Leica and the companies that use LPF is the post-processing software. In this area the Japanese companies have a major lead and in fact it is amazing what the software can accomplish. But there are limits as usual. These are not often discussed.
Above the Nyquist frequency (the safe zone so to speak), the contrast may be quite high, but the contrast is dependent on the phase of the periodic structure that is being detected by the sensor-raster. A high contrast can be found with a 'wrong' phase and will introduce artifacts in the image. The world is not so clean as we want it to be.
The Leica M solution is less dependent on the post-processing software and will allow the optical quality of the lenses to be recorded as honest as possible.
In the end both solutions may deliver the same result at the limits of resolution, but we cannot emphasize enough that these limits are not always relevant, nor defined with accurate measurements.
The test
I used for the test a Siemens star pattern with very fine structures. This pattern has a good correspondence with actual imagery and when the test pattern is well recorded, so we can confidently predict that practical results will be quite good as well.
The chart has been photographed at a distance of about two meters. At this distance the magnification of the chart will be such that all periodic structures will be recorded and more pictures are made at this distance than at infinity, where most tests are conducted. But the performance of the lens is different at infinity and at medium distance ranges. I used a variety of lenses form Leica and Zeiss. (a separate report will discuss this). For this test report I will concentrate on the new 1.4/50 asph (SLA) and the new Summicron 75 (ASC).
As comparison system I selected the Canon 5D. This has a sensor with the classical 24x36m format and delivers excellent, if not superior imagery. The lens was the macro-2.5/50mm. Form previous MTF measurements I know that this lens at medium apertures is quite good and should be a better match for the Summilux than the Canon version of the 1.4/50mm I am not interested here in comparing lenses, but comparing sensor quality. We would expect that the Leica M8 must be at least the equal of this system.
The following illustrations are enlarged sections of the sensor area. Leica recommends the use of Phase One Capture One LE software that is in the box. I compared the results of the CO LE with Photoshop CS2 9 and Camera Raw. The differences in processing are very small, so I opted for Adobe. Camera Raw settings are: sharpness: 25, colour noise reduction: 0. All adjustments are on Auto. The picture represents the center part of a much larger image file; in fact the selection is 1/40 of the total area.
Both pictures are raw versions with minimal adjustments. On screen you can detect colour artifacts in the fine details of the spokes of the star as imaged by Leica.
The Canon image too has this type of artifacts, but slightly less visible. It is now easy to jump to conclusions, but one must take into account some additional facts. This star-test is quite heavy on the image detector and the fineness of the detail is such that one will not encounter these fine details in practical work quite often. Secondly there is a difference between screen representation and print representation. I made prints of the whole test chart image with the Epson R2400, controlled by the Photoshop print options. On the A3 prints, the differences are almost negligible! We should then add the printer software as an additional element in the digital image equation. Thirdly and most significantly we should realize that at these high frequencies the results are not reliable in a consistent way and an element of chance is introduced. There is an almost universal tendency to interpret figures as absolutes and attach great value to small differences. Some magazines and also some websites present resolution figures in absolute terms like this: sensor/lens combo A has a resolution of 1836 lines per image height and combo B ' only' 1470 l/ih. It is then assumed that 1800 is better than 1500 and that these values are attainable in practical work. Both assumptions are wrong. Without the additional parameter of contrast these values are hardly insightful.
My measured results for the M8/75 combo are 1300 lp/ih (2600 l/ih) at 10% contrast. At 50% contrast the resolution is about 900 lp/ih or 50 lp/mm! The Leica claim then is substantiated? Not exactly: depending on the direction of the spokes of the Siemens star the MTF varies between 50% and 30%. Here you see clearly the danger of presenting figures without context. Based on the same measurements we can present the MTF as best value (50) or worst value (30) or average value (40). The mere figure in itself does not convey relevant information. There is also the issue of averaging the results over the image area: it makes a difference whether you present values from the center of the image or averaged over the whole sensor area. In the first case you test the maximum performance of the sensor and in the second case you present numbers that represent the average quality of the lens behavior over the image field. I have analyzed countless measurements with the M8 system and several lenses at all apertures. My view at this moment can be summarized as below.
The theoretical Nyquist limit for the M8 sensor is 73.5 lp/mm. My best results (see above) are close to that value for the Leica M8 system at aperture 4 (beyond this value the diffraction errors become visible) : about 1300 lp/ih or about 72 lp/mm. More is not attainable with this sensor. But you can infer from the pictures above, that at these extreme resolutions all kinds of artifacts are introduced. Experience tells us that good definition should be found at values that are 0.7 or 0.5 of the theoretical Nyquist value: in this case around 40 to 50 lp/mm.
When you study a large amount of MTF graphs that represent the M8 level of performance (and the same applies to the Canon set of graphs!) it is safe to say that up to 800 lp/ih the results are representative of practical performance and above 900 lp/ih we enter a zone where large margins of error must be assumed and the results around and above 1000 lp/ih are at best interpreted as tendencies.
Results compared with Canon 5D
The Canon picture presents equally good imagery, but see the comments above). At least in my lab conditions the results are comparable. But one needs additional information to see the results in perspective. I will not present the graphs as I have noticed that people will interpret the values at a level of precision that is not realistic. The Canon system has an inherent 30% advantage in imagery because of the 1.33 angle reduction for the Leica M8. In normal parlance: when we want to print an image at the same size (A3 as example) the smaller sensor needs a higher level of definition to present the same quality.
The measured result for the Canon system at aperture 5.6 to 8 is about 1200 lp/ih at contrast values from 10% to 30%. That is 50 lp/mm at quite low contrast values. But when we correct this resolution with the 1.3 factor, we can see that Leica needs 65 lp/mm to bring the same detail on the print with the same printsize. At these values the Leica has a contrast of 10%. But more meaningful is this comparison: At 1000 lp/ih Canon brings a contrast of 80% into the scene and Leica has 60% (75mm ) and 50% (50mm). All values at optimum apertures. The Canon behavior is quite intriguing. All Leica graphs follow the familiar pattern of a best value at lower frequencies and then a gentle roll off to the zero position. Canon has a different behavior: Here the values stay quite high from lower frequencies to mid frequencies (about 1000 lp/ih) where the contrast gets a software induced boost. A value of 80% contrast at frequencies around 30 to 50 lp/mm cannot be generated by the lens alone. Here the post processing must help. The upshot of the Canon approach is a high contrast image over a wide range of frequencies that is quite pleasing for the eye and also brings that digital look/fingerprint in the picture.
I am not taking sides which approach is best: the Leica pictures are more in line what you expect from silver-based negatives.
I do not have the intention to pursue the comparison Canon-Leica ad nauseam. Suffice it to say that Leica with its smaller sensor and thin glass plate presents results that up to A3 can compare quite well with the Canon images and do present a different look. If you are a number-crunching person, you might say that the Canon system has the edge even with less superior lenses.
All tests and evaluations were also done with the Summilux-M 1.4/50 ASPH. The results are not that different form the 2/75 that a whole paragraph should be devoted to this topic.
Variations in image quality when sensor speed is changed and JPEG sharpness is changed.
The same series of tests done with Raw pictures was also conducted with the various JPEG compressions and the various ISO values.
For the comparison of JPEG sharpness results I used the Summilux at f/5.6. Again there is a bandwith of values in the high-low range. At the meaningful 1000 lp/ih we see these values: Off: contrast = 5 to 10%; low: contrast = 20 to 30%; medium: contrast = 20 to 40%; Med high: contrast = 25 to 45%; high: contrast = 40 to 60%;
The difference between the Raw image and the JPEG image at medium sharpness level is not that great. In many cases one can shoot pictures confidently with JPEG without having to fear that definition will suffer. Other aspects of course underlie the usual JPEG constraints.
ISO values were compared using the 2/75. Again at 1000 lp/ih the values are ISO 160: 20 to 40%; ISO 640: 5 to 30%; ISO 1250: 5 to 20%; ISO 2500: 3 to 10%. The results indicate that (noise issues apart) ISO 1250 is the maximum speed at which you will see good contrast values to get good definition of fine detail, but is safer to set the maximum at ISO 640. I leave it to the user of the M8 to find out the best combination of ISO speed and JPEG compression to do the job as required.
The consistency of results in practical work.
Under the lab conditions it is quite easy to find the best focus of the camera/lens combo, you would expect. But it took some effort to focus accurately to be able to capture the finest possible details. I also noted that some movement of the camera (by changing the distance unintentionally or by taking handheld shots as a comparison with the tripod-based pictures) could degrade the image to some degree. Most users of the Leica M will take pictures in handheld conditions and we all know from practice that it is quite easy to move the camera when taking portraits or photographing persons in action. The lab-related pictures may show the maximum possible quality, but it would be unrealistic to assume that these results can be reproduced in normal shooting. I also became aware of the fact that when making a series of pictures where you try to find the best focus per every individual picture, you may find that some pictures are way off the mark. That is normal and again substantiates the claim that testing must be done in controlled situations to avoid all kinds of error introduced by accuracy and consistency issues.
Below you find a picture of the Siemens star with the lens wide open and deliberately defocused, but not by very much. Within the central part of the star image you see a blurred area where some patterns can be recognized. This is a phase jump of the periodic pattern and is an indication that the focus is off. The drop of contrast is most visible. The resulting picture looks similar to what you get when using a LPF before the sensor! I have presented this picture to show that there are major differences in image quality, simply by a defocus blur. On the other hand we may be happy that the mechanical precision of the Leica is so accurate that the results can be very good. But we should realize that image quality may be much lower than we anticipate: in may cases we have only our own level of competence to blame!
The picture displayed here is a really critical object: it is flat and has extremely fine lines. In real life we will take pictures of three dimensional objects with depth and then the defocus will be less visible: the sharpness plane will be at some location on the object and there is visible sharpness, when not at the intended spot.
Here again we see that trying to get consistent and repeatable results is less easy than is often assumed. Results from tests may vary according to the care and expertise of the tester.
Conclusion
The results of this test indicate that the Leica M8 can deliver results that are excellent and will not unfavorably compare to other top quality systems. Especially when one takes into account the smaller sensor area.
The Leica M digital solution does exploit the image quality of Leica M lenses to the limit of the 50/50 rule (50 lp/mm at 50% contrast). For best range finder accuracy, the use of the 1.25 magnifier is strongly advised. I have also indicated that there is a wide margin between the best possible (maximum) results and practical results. I did note that the mirror movement of the Canon could destroy the delicate fine lines of the target, even when the camera was on tripod.
One of the issues I wish to ask attention for is the focus topic. With digital imagery, the cost of focus bracketing is zero and it will definitely help to get better imagery in many situations. When using the 2/75 with focus bracketing I noted that the steepness of the focusing curve in the lens mount implies a definite on-off position for sharpness. That is fine for quick and accurate focusing, but the steepness makes it difficult to accomplish fine tuning in the focus bracketing mode.
For best definition we should employ Raw images and stay within the ISO range from 160 to 640.
Preview
In the next part I will compare the M8 images with the M7 images made with the same lenses. Can film go further than the 50/50 rule states?
In addition I will compare pictures of 3-D objects to see whether the conclusions from this lab report need fine tuning for real life situations.
I will also look at the dynamic range, the noise and the color aspects of the M8 and discuss the IR-issue, that has popped up in the internet.