Sunday, March 27, 2016

Resolution of historic photographs in examples, part 1

As everyone knows, historic photographs are full of lovely and important details. Well, at least everyone who had paid enough attention to explore an original or had chance to see a quality scan. In 19th century photographs was not enlarged and the prints matched the size of negatives. Even small CDVs was intended to be viewed with care under a magnifying glass.

The estimate of upper bound for the resolution of historic photographs is essential for setting up the parameters of digitization. Today we finally have technology to digitize practically all the picture information contained in the photograph and make them widely available to public. Digitization is laborious and dangerous to the original and thus it is desirable to do it just once. The key question thus is what resolution of scan is enough so we will never need to repeat it again? I asked this question in 2004 when our project of digitizing the archive of Šechtl & Voseček started and here I try to summarize my observations about one particular image.

Update: See also part 2.

The microscopic strike in Bosnia

Ignác Šechtl: Žižka Square in Tábor, 1880-1885, wet process negative 11x13cm,
http://sechtl-vosecek.ucw.cz/cml/desky/deska0090.html
This is one of my favorite examples of a high resolution historic photograph. It was one of first photographs we scanned in 2004 and noticed something extra-ordinary. A tiny crop of the original glass plate shows clearly visible text "Povstání v Bosně"/"Strike in Bosnia"! This was the key to date the photograph - the strike in Bosnia was in 1875, so this must be one of oldest preserved photographs of the square!

We had to wait another 10 years to read more of the text, because you need incredible 5588DPI scan which is presented here in full resolution. In 2010 we purchased Eversmart Select scanner capable of doing that (this king of the flatbed scanning used to cost like a Ferrari in 2000s and we still paid $6000 for that monster). The text itself is 0.5x0.2mm in the original size. Click on the image to see the full resolution
Image inverted to positive and tiny crop (4.5x5.5mm) shows people posing for the long exposure but also well readable texts in the news stands.
The photo contains over 10 people posing in the back of the square. The boy in the front (outside of the intended crop for CDV card) managed to get himself photographed 4 times. Clearly the exposure time was long and the whole square was cooperating with the photographer. Each person is about 0.2mm tall on final photograph. They may have been disappointed to not be able to recognize themselves even with a strong magnifying glass.

We however were able to read more of the text. You can see "Národní Kalendář"/"National Calendar" and even smaller type "výročí"/"anniversary". With help of Google I was able to identify the calendar itself it is "Pečírkův Národní Kalendář". Here you can read a thesis about this famous callendar. Clearly it is one of issues published for anniversary of the strike in Bosnia. By other attributes seen in the photo we know it was taken in 1880-1885. It is thus matter of the visit in the national library to work out which one it was.

So what is the resolution of this photograph?
Crop of the advertisement scanned at 5588DPI using Eversmart Select (left), after digital sharpening (middle) and after downscalling to 2000DPI and resizing back
It is possible to see that 2000DPI scan would not capture all the detail of the smaller text. At 2000DPI the scan of this image would be 8701x10315pixels, or 90 megapixels! Well above resolution of my Canon G7X and this is quite conservative estimate.

Of course the photograph is not on par with today 90 megapixel digital camera across the whole plate. Historic lenses was sharp and contrasty in the center and the resolving power degrades around the corners. Still it is impressive that in 1880s photographers was able to meet resolution of expensive professional cameras available today. 80 megapixel Mamiya sells for 28000USD and that is one of cheaper options.

How can I estimate the resolution of historic negative?

Is my 5588DPI scan enough or can I read more of the advertisement? Well, in this case I can answer easily by comparing the sharpness of the crack in emulsion with the image details. Clearly the crack is sharper than rest of the image and thus my great-great-grandfather's camera did not capture sharp details at 5588DPI.

Given large archive of negatives one however can not explore each in this detail. Fortunately there are measurements available which let us to give reasonable upper bound on the image. They are well summaries in work of Timothy Vitale: Estimating the Resolution of Historic Film Images: Using the Resolving Power Equation (RPE) and Estimates of Lens Quality.


Since 1940s some film manufacturer publish data on number of line pairs per millimeter which can be captured by a given film. By summarizing data from published measurements T.J. Vitale prepared the following table on B&W film resolution:
Native resolution of film.
Timothy Vitale: Estimating the Resolution of Historic Film Images: Using the Resolving Power Equation (RPE) and Estimates of Lens Quality, Figure 18a
This table shows data on resolution of B&W film (starting in 1940) and a predicted resolution of the negative based on year of manufacture.

The resolution of negative is however just very crude upper bound on resolution of the actual photograph. Every photograph is taken through lens and it depends on the quality of lens how fine details they can capture. The following image nicely summarizes the effect:
Graphics by Norman Koren explaning loss of resolving power by combination of film and lens.
http://www.normankoren.com/Tutorials/MTF.html

To estimate the resolving power of the photograph one thus needs to estimate the resolving power of the lens and combine the results using resolving power equation or use the following simplified table predicting the loss of resolution:

No.% loss Description of Historic Era, Film Format and Lens Quality
125 modern medium resolution film in 35-mm & 2¼”x2¼”, thru an excellent lens (100 lp/mm)
240 modern small format film thru an average good lens (80 lp/mm) with good processing
340 average small-format 1940-70 film exposed through excellent lenses
440-60 large format film (1890-1970) thru average quality (40-lp/mm) lens with fair processing
550-60 modern large format film exposed through a good quality (60-lp/mm) large format lens
650-70 very high resolution film thru an excellent lens (100 lp/mm)
755-70 modern high resolution (5000-7000 ppi) film exposed through a good lens (80 lp/mm)
860-70 all common modern film through an average (40 lp/mm) quality lens
960 large format film (including early roll film) from 1890-1930 through average quality lens
1060-70 very early film (1890-1930) thru all possible lenses, assuming good alignment and focus
1160-80 large format film and glass plates 1875-1900 through average LF lenses (10-20 lp/mm)
1260-90 modern large format film exposed through older lenses or average large format lenses
Twelve Guidelines for predicting percent film resolution loss due to lens era and film format.
Timothy Vitale: Estimating the Resolution of Historic Film Images: Using the Resolving Power Equation (RPE) and Estimates of Lens Quality.Table 5
To determine a resolution of the black and white negative one first estimates the resolution of the film material using the first table and then account loss using the second table. To be sure that picture detail can be distinguished from damage, it is good to scan at about a double of the resolution obtained.

So what scan I need for my great-great-grandfather's photo? Well it is large format negative taken in 1880's. According to the first table the resolution of the "film" should be 1500PPI. I apply guideline 9 to an estimated 60% loss. The photograph thous should be at most 600PPI. In fact T. J. Vitale would rank it even lower:
1889 Film - The native resolution of film from 1885 would be about 1460 ppi (28-lp/mm). This is a low-resolution film exposed through an average lens of the era, about 20-30-lp/mm. Using Guideline 10 in Table 6, the film would produce images with resolution in the range of 483 ppi, as shown in column 7, in Table 7. However, if the image was made with a hand-held camera such as a Kodak #3 Folding camera or the Kodak #2 Box camera, it could be even as low as 292 ppi, following Guideline
11, as shown in column 8 in Table 7.

Timothy Vitale: Estimating the Resolution of Historic Film Images: Using the Resolving Power Equation (RPE) and Estimates of Lens Quality, page 22

This is what I should see in the news stand if resolving power was 600DPI:
The same crop downscaled to 600DPI and sclaed up.
How did my great-great-grandfather cheated the technology and obtained such a great resolution at a time it should have been impossible?

First of all I believe there is a mistake in Timothy's Vitale's table. 1940 is about the time when enlargers became mainstream (with advent of 35mm cameras) and photographers moved away from contact copies. For the first time the grain of negative material started to matter. I thus believe that the use of Moore's law is wrong. At the time black and white negatives was invented, they started with some resolution which was easy to produce and the improvements before 1940s was probably much less dramatic than improvements after 1940s till present day.

We can thus assume that the film resolution remained constant from the time film was invented to 1940. Lets repeat the estimate using data from 1940. 1940s film is rated to 3000PPI, with 60% loss I get 1200PPI.

Crop scaled to 1200DPI and back to 5588DPI
Well, this is closer, but clearly I would not be able to identify the author of the calendar. So still not good enough.

There is another mistake in my estimate - the photograph is taken by very laborious wet collodion process (used by my great-great-grandfather till early 1890s), where the glass plate had to be prepared on the spot, exposed and developed while still wet (so photographer had to carry a tent with mobile darkroom). Wet process was well received for its low grain and used for reproduction long after the dry negatives emerged.

Lets thus assume that the negative has infinite resolution and look into what lenses was available in 1880s.

DateCause of ImprovementProfessional Large Format in lp/mmAmateur - Box Folding & PnS in lp/mmProfessional Medium Format in lp/mmPro & Amateur Small Format lp/mm
1826base line<20NA NANA
1835Chevalier Achromat20ishNANANA
1841Petzval Achromat20-30NANA NA
1873Abbe Optics 20-40NA NA NA
1886Zeiss Apochromatic 30-40 <20NANA
1893Goerz Dagor Achromat40-6020-40NANA
1902Tessar hi-contrast40-6020-40NA NA
1925Leica RF/FPS Elmar 40-60 20-40NA 50-70
1929Rolleiflex MF Zeiss40-6020-4040-6050-70
1935-40optical coating 40-70 20-4050-7050-80
1948Hasselblad MF Ektar40-7020-4050-8050-60
1949-59first SLRs - C, N & Z 40-7020-4060-100 40-80
1960-70adv lens coatings40-8020-4070-10040-100
1970cheaper optics 40-8020-4070-10040-100
1975-88LF lens coating 40-9020-4070-10040-100
1987point-n-shoot40-9020-4070-100 40-100
Lens Resolution Estimator.
Timothy Vitale: Estimating the Resolution of Historic Film Images: Using the Resolving Power Equation (RPE) and Estimates of Lens Quality.
By this data in early 1880s the best available optics was sporting 40lp/mm max. The digital resolution can be easily computed by 40*25.4*2=2002 (conversion to inches and multiplication by two for every line pair). This is a plausible number matching my observations. But where is the loss for film resolution? Even if fine grained, wet collodion process was definitely not infinitely resolving. Well, I was not able to find any data and I am considering to do an experiment.

In every case, the resolving power of lens should be at least 25% more than the actual resolution of the image. It still seem unlikely Ignác was able to reach this resolution with contemporary camera, bcause lenses resolving over 50lp/mm became available only in 1890s after the houses photographed has been demolished.

By family legends Ignác traveled to England and Hungary to obtain quality lenses. Perhaps he also traveled in time for his shopping?
My great-great grandfather visiting London.
The more likely explanation is in the way lens resolution is rated. The resolution differ at the center and towards the edges and is also greatly affected by the width of the aperture. In the optimal spot and with small aperture the lens can probably perform slightly above its official parameters even though T.J. Vitale explicitly mentions historic lenses (1915-50) to be limited to 40-60lp/mm in the center and the negatives not exceeding 1250PPI in his experience.

Digitization guidelines

What are the official guidelines for digitizing glass plate negatives? The answer varies but mostly it is well bellow my observations:

Summary


Digitization of historic photographs should be done in a way that all details of the originals are captured in the digital file. Here I focused on the black and white negatives which are among the finest and also show the largest variety. The resolution of photograph can be estimated based on year it was produced and the size of negative using data presented by T. J. Vitale's work.

I can show a real example of photograph taken 1880-1885 which rivals best digital pictures we can take today (at least in the very center). Image details reaches resolution over 2000PPI and needs scan at over 90 megapixels. I would say that really fully satisfactory scan of the presented photograph (in the terms of resolution) should be at least 4000PPI, or  360 megapixel. Doubling actual image resolution is useful to make photograph damage easily distinguishable from original image (in fact, T.J. Vitale recommends to multiply by 4 for perfect capture, this is supported by Nyquist theorem). Quite a monster which exceeds most of the recommendations of the digitization guidelines with exception of FADGI.

Following guideline of T.J. Vitale, one gets estimate of 600PPI. The prediction of film resolution before 1940s, presented by T.J. Vitale, should thus be taken with a care and probably one should stick to the published data (that is, all dry plate negatives before 1940s should be believed to resolve approximately 3000PPI) and treat wet process negatives even more carefully. Negatives predating 1940 are rare and precious and the extra disk storage for a good scan is worth it.

Next I will give examples of high resolution dry process negatives from our collection and move on to the question of dynamic range, bit depths and other aspects of a good scan.

Update: This blog post describes very similar experience.

2 comments: