This article describes, how I made the resolution-power of lenses digitally measurable on analog film and COMPARABLE to the data, which are directly measured on digital sensors.
Since a long time I am looking for an experimental set-up, which allows me to understand, how the information content of the exposure on an analog film compares to the digital data from a digital sensor – looking through the same lens. Resolution being the main point of interest for me in this case.
FAZIT of this article is: I just found a method to do this … and it works!
Fig. 1 is an example of the resolution-chart, generated with Olympus OM Zuiko Auto-W 28mm f/2.8 lens on black and white negative film (Agfa APX100) at the lenses optimum aperture f/5.6. It shows the MTF30-resolution-values from center to corner (21,7 mm picture-circle-radius):
1. Introduction: Quality of historical lenses.
Until the end of the last milennium, photographic lenses were designed to generate highly detailed images on film (or glass-plates coated with light sensitive layers).
When at the beginning of the millennium 2000 ff the radical change to digital sensors for the recording of photographic pictures happened within a few years, the photographic community sat on a huge number of phantastic and beloved lenses for the ancient analog recording method with film. We all could not believe, that all these wonderful and iconic lenses would become useless.
This was a strong motivation, to put the historical lenses in front of the sensor-driven digicams! Knowing very well …
… that for the very special needs of digital imaging sensors, new lenses had to be designed under consideration of some special optical features along the optical path behind the lens, before the light hits the light sensitive elements.
The older „analog“ glass is not perfectly fitting to these conditions! … but what are the artefacts on sensors and how important are they?
Measuring the Modulation Transfer Funktion (MTF) of a lens is a fundamental method to characterize the optical resolution of a lens itself – independant from the recording method (film or digital sensors): the direct optical measurement of the MTF-curve over the imaging circle of a lens shows the contrast as a function of the „spatial frequency“– these data today often being available from the lens-makers. Historically, the method was developed parallel to each other by Zeiss and Angénieux (during the 2nd world-war). Fig. 1 shows the certificate, which was delivered with my Angénieux zoom-lens as a qualitiy-proof (showing that this is an extraordinary example … above average expectations!).
Reminder: One „cycle/mm“ is two lines/mm! Two lines (one black + one white!) are the spatial (linear) cycle. We call this here in my article „Linepairs = LP„. In the systematics, which I personally use in the IMATEST-software, the resolution is always defined as „Linepairs per Picture Height = LP/PH„, where the picture height on film or sensor always is 24mm (for 35mm stills) … if not otherways stated. That means:
For 35mm stills (24mm x 36mm) the data in „LP/PH“ can be converted into „cycle/mm“ with division by 24, or into lines/mm with division by 12.
Fig. 2 shows the type of MTF-plots, which are today typically used as the contrast over the radius of the image-circle:
These lens-specific-informations are perfect to characterize the lens – but do not help me, in the case of historical lenses, for which these data are mostly not available. For these I have to measure the performance myself – either using analog film or the digital sensor.
In historical times (1960s to at least 90s) tests of „Lens Performance“ were published in several US-publications like „Modern Photography“ and „Popular Photography“ and others targeting on amateur photographers.
These mostly used the „USAF 1951“ target, which was mounted on a flat wall and photographed at scale 50:1 on b&w negative film and read by inspection with a loupe of defined enlarging power directly on the film strip,
Results from these test were as shown in the following picture:
I tried this method myself in the 1970s and can tell you: it’s a lot of work!
2. The digital IMATEST lens testing method and software for digital imaging:
Testing the optical performance on a digital sensor is facing several facts and influences, which are new and specific: pixel size, algorithm, problems of digital signal-processing systems like aliasing, additional optical elements in the optical path like glass-filters and micro-lenses!
The question: is there an essential influence of all these optical systems on the visual result in the picture over the picture-circle (Bildkreis), e.g. because of the varying angles at which the light-rays hit on the sensors between center and the farthest corner of the picture format or due to the additional optical elements introduced into the light-path?
In the case of RANGEFINDER-lenses we know that there often is a strong influence of this. These lenses are often made for a very short distances between the last lens and the film – especially for wideangle- and standard-lenses. Little was known to me about historical SLR-lenses, which were never planned and calculated for the use with modern digital sensors.
Since several years I do quite a few measurements on historical lenses, using a high-resolution digital sensor with 62,5 Mega-Pixels, resulting in 61 MP effectively on Full Format (35mm stills).
Until now I did not know, whether the measurement of my historical lens is falsified due to artefacts, genrated by the digital recording system. The work, described in this article, was done, to clearify this situation.
I just want to know: how does picture quality of historical SLR-lenses on the analog film compare measurably to that delivered by digital sensors?
Digital cameras are really big number-crunching-machines! And with the right software, I can use the numbers to generate a numerical picture of the optical quality of the lens-sensor-combination. IMATEST is such a software and it uses standardised TARGETS to do that. I use the following target:
Over years I did – like many other amateur-photographers – compare real-world photos of analog vs. digital processing. But I was never satisfied, because this method gave me only subjective impressions – it did not create reproducible figures, to generate a precise description of the results!
I collected intensive experience with IMATEST on more than 150 lenses over meanwhile 5-6 years using the digital pictures generated by digital sensors (4,9 to 102 Megapixels) of seven different DIGICAMS. During this time, my Standard Digicam to compare lenses was (and still is) Sony A7R4 (60 Megapixels) – since it had arrived in the market (2018/19).
IMATEST (Studio) software delivers MTF-based resolution data – as it can do that separately in three RGB-channels, it also delivers lateral CA-data. Using the Target structure of Fig. 5, the software selects 46 local areas, and runs the MTF-measurement automatically for all these 46 areas. The following picture demonstrates the automatic areas, which are typically selected – but you could choose others as well:
These are the curves, which are generated from each digital picture (black&white):
The upper left curve shows the edge-profile at center of the target (ROI no. 1, which is the left (vertical) edge of the center square in Fig. 6). From this graph the edge-rise between 10% and 90% is taken from the x-coordinate in pixels. The lower left curve is the MTF-curve (contrast over spatial frequency) for the same location. From this graph the MTF30 value (Frequency at 30% contrast) is taken: follow the horizontal line at 0,3 MTF-value to its section with the curve and take the frequency on the abscissa. The right curve shows the MTF30-values of ALL 46 ROIs plotted over the distance from the center in the 35mm-fframe.
I have resumed the IMATEST test-method in more detail in this article here in my blog!
3. Digital measurement of resolution on analog film
Now I decided to make the following experiment:
- Take a photograph of the IMATEST-target on analog film;
- digitize the picture with a film-scanner;
- analyse the resulting digital picture with IMATEST.
For the tests, which I describe here, I used the following hardware:
- Camera for the shooting on analog-film: Olympus OM-4Ti
- Lens: Olympus Zuiko Auto-W 28mm f/2.8 (Ser.no. 232073)
- Film: B&W negative film AgfaPhoto APX100, iso100, developed in Rodinal 1+25 (8′)
- Scanner: reflecta RPS 10M film scanner
The OM-4Ti (about 25 years old) and the lens (nearly 50 years old) work still perfect. I let the OM-4Ti automatically generate the exposure time: from 0.4 seconds to 1/250 seconds. The densitiy of the negatives was very constant on the film-strip!
Why I chose the OM 28mm f/2.8 I will be explained in the next article „My crazy lenses – One Lens + Seven Cameras: Olympus OM 28mm f/2.8 – Analog vs Digital – Part I: Analog“ …
With this method I hope to use the full analyzing-power of IMATEST-software on a picture-frame, which is generated through the lens WITHOUT the typical artefacts, which digital sensors MAY generate in the optical path of a historical lens.
ON THE FILM we have now the IMATEST target-pattern, which allows to make a fast and powerfull analysis of optical data over the full picture frame – very close to the edges and into the corners. This pattern is superimposed by the typical grain-structure of the light sensitive layer – and potential light-diffusion-effects within the film thickness. Both (analog) effects LIMIT the resolution, which can be achieved on FILM.
My first and major interest was always focused on the observation of the enormous difference between the center-resolution (see Fig. 7), which is digitally measured on A7R4 with ca. 2,800 LP/PH or higher) and corner-resolutions of 400-600 LP/PH on the sensor .
With other words: are the low values on edges and in coners of the frame, measured with the digital sensors, an artefact, caused by the change of angle, under which the light-rays are hitting the sensor (or by the filter stack)? We know definitely about these effects with rangefinder-lenses, which have a very short back-distance between last lens and film, causing big trouble on sensors of mirrorless cameras. This is today well known, to be mainly caused by the thick filter-stacks in front of the sensors (creating field-curvature and cromatic aberrations with analog lenses).
It has been shown, that this can mostly be „cured“, by reducing or deleting the filter-stack, and/or putting a positive lens (so-called „PCX-filter“) in front of the lens-sensor-combination.
I made my first attempts to photograph the IMATEST-target on film parallel with
- b&w-film Agfa APX100, iso 100 and
- colour negative film EKTAR 100,
which are both still available as „fresh“ products.
I will report here mostly about the black-and-white film results, because the results on colour film were very low in resolution – in fact only about 2/3 of that with the b&w-film. The colour-neg-film had been externally processed by my photo-dealers lab – while I did the devellopment of the b&w-film myself with Rodinal. As a consequence, my confidence into the colour-results is not too high … it has to be reviewed definitely! The lateral CA-values, however, were very well on par with the measurements with the sensor. I will re-try! Maybe with a different color film.
The negatives were digitized through my film-scanner reflecta RPS 10M,which offers a maximum linear resolution of 10,000 pixel per inch (PPI).
To me, this step seemed to be very important: to avoid new artefacts from the digitizing algorithm. So I chose a spatial frequency, which is higher than the expected limiting spatial frequency of the film: I set the scanner at 5,000 ppi. On pixel-level this corresponds to an imaging-sensor of ca. 33.7 MP (for 24mm x 36mm) -digitaly speaking it is an „8k-resolution“.
From my earlier estimations I had found, that a normal recording film for general imaging purposes should correspond to a digital FullFormat-sensor with 20-12 MP.
The picture height, which the scanner digitally delivers (24mm minus a bit of crop to frame the target safely), was 4,536 pixels and so the „Nyquist Frequency“ of the scanner set-up corresponds to 2,268 LP/PH – corresponding to an effective sensor-size of 30,3 Mpxls.
Fig. 9 shows the b&w-picture, which was generated with the scanner:
Let’s have a closer look into the structure of this image – in Fig. 9a you get an impression of the grain structure of the films emulsion at about 200% enlargement of the 30,3 MP-image:
Following picture is the MTF-curve of the image „as scanned“ (in the center of frame):
The „noise“ in the curve is caused by the film-grain, which is considerably bigger than the pixels.
I transformed these image data to a picture height of 3,024 pixels – of course by proportionally shrinking the pixel-count in the width as well.
Previous trials had shown, that with a film with this grain-structure, this digital image-size would give adequate results for MTF and resolution, what the following summary of the IMATEST-Data confirms:
With a pixel size of 7.9 µm we got a much smoother MTF-curve. The Nyquist-frequency of 1,512 LP/PH would allow to measure quite a bit higher MTF30-resolutions. On the same APX100-film-strip I had run the same test for three other lenses of the OM-system. In the following spreadsheet you see the highest MTF30-resolutions CENTER and CORNER (mean values) measured for for all three lenses:
Looking back to the results, which „Modern Photography“ and „Popular Photography“ published in analog lens tests of the 70s, 80s and 90s, measured on film (Plus-X from Kodak) and analysed visually with a loupe, we see, that the resolution-readings in the corners (sometimes called „far edge“ for the short edge of the frame) often started at 45/50 lines/mm with widangles and reached 75-90 lines/mm for the best lenses. This seems to be in fair correlation with the results of my method.
My interpretation of what we see here:
I. Looking first at the resolution-values in the center of the measurement on film we see something like a „barrier“ in the range of 1,350 LP/PH looking at different top-notch Olympus lenses.
With the A7R4-Sensor we measure two and a half times the resolution (2,767 LP/PH) compared to that on film (1,099 LP/PH) in the center of the OM 28f2.8 – at open aperture (f/2.8)! (See Fig. 7 / Fig. 11)
Conclusion 1: the maximum resolution-values, which we see with this method here on analog film are limited by the GRAIN of the film – and here they correspond to ca. 112 lines/mm. I estimate the MTF30-resolution-limit of the APX100 film to be 120 lines/mm respectively 1,440 LP/PH. For the (old?) Agfapan APX100 Professional film a resolution value of 100 lines/mm at 30% transfer factor value was given. (I do not know, wether the new APX100 emulsion (ADOX) matches with the former product exactly. Of course, the chart contrast of the used target has also influence – in my Imatest target it is 4:1.)
The (relatively small) differences with different lenses are probably caused by differences in contrast of each lens – which is also the reason, why in most cases the center-resolution rises by 10-20% from open aperture to f/5,6 and then drops again slightly. The width of the edge-profile in the center is also pretty constant (0.011-0.016 mm), which probably corresponds to the average film-grain diameter.
II. Looking at the corners we see the maximum corner-resolution on film typically lowered by 12-23% compared to the maximum center-resolution for these lenses – in the case of the OM 28 f/2.8 at f/2.8 the relative corner-drop (from 1,099 LP/PH at center to 734 LP/PH in corners) is exactly 33%. (see Fig. 9). The lowest corner values thus correspond to 50% of the grain-determined resolution-limit of this b&w-film.
Conclusion 2: The MTF30-corner-resolutions, measured with Imatest on the b&w-film APX100, delivers the REAL resolution values on film, because these values are far enough away from the resolution limit of the film, to be un-influenced by the grainyness!
Now we look at the results measured on digital sensor and on analog film, put together in one graph:
Fig. 13 shows principally the same Data, which were the basis of the graphs in Fig. 7 and Fig. 11 – right side (MTF over disxtance from center). Here the mean values of the MTF30-resolutions (center – part way – corner) are plotted over the distance from the center of the image.
The horizontal lines in Yellow and Green represent the Nyquist frequencies of the A7R4-sensor and the APX100-film respectively.
The blue line stands for the MTF30-resolution on film (734 LP/PH in corners).
The grey line shows the MTF30-resolution on the A7R4-sensor (784 LP/PH in corners).
Final CONCLUSION: yes, the enormous drop of resolution over the distance from the center, which is measured with high-resolution sensors, is real. This is no artefact, generated by the sensor in the corners but corresponds quite well with the resolution measured directly on film. Conclusion 1 has already shown, that the high resolutions, measured in the center of the image, are also real and the resolution-results on film are just limited by the grain of the film.
This seems to be widely valid for many historical SLR-lenses, which have an SLR-typical large back-focus. As we know, with RANGEFINDER-lenses having very short back focus to film, this may change dramatically – and that is covered widely in literature in the meantime. (Keywords: Sensor filter-stack; PCX-filter)
I have myself to-date researched the phenomenon on the example of the rangefinder-lens (LM-mount) Voigtländer VM Ultron 35mm f/1.7 in this article. Unfortunately until now available only in German …
My intention was, to find out, whether this corner-degradation-effect would also influence the picture quality with SLR-lenses:
… well – it obviously is NOT, even with wideangle lenses!
Another conclusion, which I draw from these results is, that the calibration of the film-scanner-resolution, set to 1.512 LP/PH (Picture-height of 3,024 Pixels), was a realistic choice for this type of b&w-film. It corresponds to a image-sensor with 13.5 Megapixels. That may need to be re-tuned with analog films with finer grain and higher resolution.
If we would use an ultra-fine-grain analog film (which exists!), we probably will measure very similar resolution curves as we get from the sensors! .. and I will definitely try to prove this hypothesis as well. (I have already tried to to this with the ADOX CMS 20 II – however, the development seems to be absolutly critical – I have to retry …)
–> It will also be interesting to extend this method an different analog films – and using the same lens for that, it may allow to characterize the film-emulsion with respect to its practical resolution power … of course with a specific calibration of the scan-process adapted to each film.
–> Last thought: I will try to find out, whether an innovative film pressure-plate, as it is used in the Contax RTS III (with vacuum!), can improve the variations of resolution, which are measured over the image frame …
At the end, you see in Fig. 14 + 15 the full resolution analysis of OM Zuiko Auto-W 28mm f/2.8 on 60 MP sensor and on b&w-film:
Finally, I want to tell you tell you a small anecdotal story about the early times of digital photography:
My own analog photograpy started in the mid-1950s (no error!) – and I was a busy hobby photographer, since then also developing my B&W-stuff on my own. So also my wife had a long pre-experience in seeing my film-enlargements on paper in sizes up to 50 cm x 70 cm.
When I made my first experiences with digital photography around 2002 with 6 MP (Canon EOS 10d), my wife was very critical about the results: „This does not look natural!“ she said. And she was right …
As a long-time enthusiast of the Olympus OM-System of 1972 I had realized of course, when Olympus tried something completely new in 2003 with the „E-1“ – a DSLR with the FourThirds format (17,3×13 mm). It had only 4,9 MP – less than the Canon, which I owned at that time. But surprise!: the pictures found grace in my wifes eyes … and my own! Since then I know: it takes more than just Megapixels, to tango!
Herbert Börger, Berlin, April 2023
9 Gedanken zu „Resolution of Lenses (Sharpness) on Analog Film – Compared to Digital Sensors“
Thank you very much for your very informative profound testing, the nice comments and lovely sidemarks.
The finding with the equal corner resolution in film and sensor is great and helps me a lot to rethink my challenge with the Contax RF wide and superwide angle lenses on a sensor camera.
I am happy, that you like my article – good luck with the Contax lenses. Is it for the Contax G1/G2? To date there is only one post from me regarding the use of RF-lenses with sensors: https://fotosaurier.de/?p=2632
related to the use of „PCX-filters“ … more to come! Unfortunately this is only available in German now.
The Contax G1 with 16, 28 and 90mm lenses had been my compact and lightweight set for hiking and travelling. According to the findings of HARUHIKO T, I‘ve adapted the wa lenses 16, 21, and 28 with PCX-correction filters (Optosigma 21/28 and Zeiss for the Hologon) to the Sony A7Rlll.
(Samples with the Hologon: https://flickr.com/photos/147484618@N03/sets/72177720296572255)
Do you use a gradient center filter? Which adapter du you use with the A7RIII? On the Hologon: which focal length has the PCX-lens? Which PCX-FocalLengths do you use for the 21/28mm ? How did you manage the focussing to infinite? I think you had to modify the adapter?
I have the G2 and will a.s.a.p. photograph the IMATEST target pattern with the 16/21/28mm lenses on analog film and publish the resolution results. I have Optosigma-lenses with several different focal lengths, too.
The supplied gradient filter I’ve not used with this samples. Furthermore it does not compensate the total light falloff of the lens and the sensor(!). I do practice 2 or more exposures with different settings and combine them as HDR using the pp software. The 70mm PCX lens (the thinnest one from Zeiss [with a little logo ‚CZ‘ on the coating], which my optician could order) has 1333mm. That was a lot less expensive than a single custom made PCX lens from Optosigma.
The Hologon lens will render the infinite details reasonable good and Zeiss informed about this lens, that the setting 1.5 mtr on the focusing scale should render infinity fine. For this lens a friend did a black filterholder in the manner of the gradient filterholder with the 3D-printer. It’s still not finished, because some optimizing in some details have to be made. For me, the PCX lens from CZ on the Hologon does a better job than the PCX lenses (1500mm) on my 21 and 28. But I may be wrong, because I’ve not done yet a fair comparison. The camera with the Hologon I use on the tripod. The 21 and 28 with the camera handheld. Also bracketing I have not yet made with these lenses. The cg-next adapters I have are those cheap ones. The look all the same but are different. One has a incorrect back focal plane (close focus only), one does not fix the lens bajonet perfectly and the one which is in use with the Hologon is very tight and it hardly can be removed(¿?¿). Better cg-nex adapters are on my wish list;)
The filter holder https://www.flickr.com/gp/147484618@N03/1c3f46PV3Y
Great! I think I need my own 3D-printer, too!
Regards – Herbert
Another question: how did you determine the focal lengths of your PCX-filters. Link?
There is a very long thread and lot of information at Fred Miranda about this sensor stack issue. There and from another site, beside the information of HarihukoT at flickr, I‘ve got the information for the 21 and 28 Biogon (sorry, no links saved).
With the Hologon and the standard Sony sensor, the limit is 1500mm for the correction filter. If some work on the CG-nex adapter in order to optimize the ‚Auflagemass‘ is done or the lens will be adapted with a smaller diaphragm(f16…f22) than there are more options. The decision to go with 1333mm for the Hologon was not bad for me. If my memory is correct, 1200mm should be optimal for Biogon 28 and 21, when using the Sony A7 standard sensor. 1500mm are the stock PCX filters available at Optosigma and they are recommended. If using the full resolution and 100% magnification, you will see probably the details in infinity, beside other things, not perfect. For me a 24mp (+/-) image is fine for Hologon exposures. Probably there are other ways too for optimization, e.g. focus stacking.
Have a look at Haruhiko T (sorry for misspelling his name)