Humans being confounded by color photography

https://www.laphamsquarterly.org/roundtable/new-look-same-great-look

by apollinaire • 4 years ago

66 18 4 years ago

retrac• 4 years ago

I am reminded of panchromatic film. The light sensitivity of silver salts on their own doesn't mimic the human eye, with the peak around the violet and extending into the ultraviolet, with lessened sensitivity in the red, down into the infrared. The result [1] (note the UK flag) is unlike what the eye sees. Panchromatic film adds sensitizer dyes similar to in colour film, to boost red response. All non-specialist B&W films are panchromatic. IR and UV filters might be necessary, too. Colour accuracy is important even in B&W photography.

[1] https://upload.wikimedia.org/wikipedia/commons/3/36/Northern...

dehrmann• 4 years ago

Photoshop (and others, I'm sure) allow you to convert an image to black and white, but control how the colors going in contribute to the output. I wonder if trained black and white photographers would be able to recognize when this was taken advantage of in non-obvious ways.

Dave_Rosenthal• 4 years ago

Yeah, a dark sky in a B&W photo is one of the most obvious ones (red filter). The other one that jumps to mind is white/bright foliage (infrared filter).

kqr• 4 years ago

This is also why many Northern Europeans in old photographs look so ghostly – the chemistry is really sensitive to the blue of the irises in their eyes so they turn nearly as white as the eyeballs.

hilbert42• 4 years ago

"Colour accuracy is important even in B&W photography."

I would argue that it is not only important but also essential. Having a lifelong interest in photography as well as having been professionally involved with electronic imaging, one of my interests is historic photographs especially those from the 19th Century—some of which are of a remarkably high quality.

The trouble is that until very late in the Century all practical photographic emulsions were only blue-sensitive. This leads to considerable distortion of the grey scale rendering and such which often gives these images a harsh high contrast appearance, especially so people's faces. Unless the photographer intervenes manually, often the consequences are that a pale light face will appear swarthy or that a person's lips will appear overly dark or even black due to the emulsion lacking a response in the red end of the spectrum.

It's best to illustrate this by example. Some of the very best examples of old photographs that are readily accessible are the Library of Congress's (LOC) Civil War collection of glass negatives and prints: https://www.loc.gov/pictures/collection/cwp/. LOC has done an excellent job in digitizing these images, it has scanned them in high resolution that would be close to the Nyquist limit and attention has been paid to capturing close as is practical the full dynamic range (grey scale) of the images. Moreover, the photographs are easy to experiment with as LOC makes available high quality images for download, the highest resolution images are in TIF format and often exceed well over 100MB in size. If one wants to carefully study 'color' distortion in B&W photographs made from emulsions that only have blue sensitivity then I doubt if there is any better place than to look here.

Many of these photographs were taken by the best photographers of their day—Mathew Brady et al, and—considering the era—many of them are of extremely high resolution due to a combination of the thinness of the photographic emulsion—a characteristic of the wet collodion process used back then (thin emulsions have low light dispersion)—and the very large size of the photographic plates which were typical of the day. The sheer sharpness of many of these photographs means that it's easy to examine both humans and inanimate objects in ways that we rarely do. A 100MB-plus TIF image allows one to zoom down onto objects that were never intended to be a principal part of the photograph. The high resolution of these images allows for a close in situ examination† of mundane objects of the era that otherwise wouldn't have been possible, and that's a big boon for cultural historians, anthropologists and the curious.

(Before commenting further, everything I mention below assumes the following caveats: that altering an image in any way renders it a fake, therefore it is critically important that all original prints, glass plates, etc. be properly archived together with precision uncorrected backup scans. This is just common sense and ought to apply to all originals.)

As implied above, there are several issues with these B&W photographs that arise from the significant 'colour' distortion caused by collodion's blue-sensitivity, and we need to know the extent of the problem if we are to attempt to extract any additional useful information from them—not to mention altering them for better appearance.

So what are the issues? First, we do not know the colour of many of the objects in these photographs let alone the accuracy of the colour—and for historical/archaeological reasons we may want to know this information (this would be important if genuine copies objects from the period no longer exist). Second, many of these photographs are of great historical importance, so we would want to present them in ways that make them more pleasant to look at—such as reducing their high contrast, stretching detail in black regions, rendering people's faces more lifelike, and so on.

I consider it important that when we attempt to improve these images that it be done on a scientific basis and not just a patch-up job in Photoshop (although I've been guilty of that in the absence of better techniques). Clearly, I cannot cover this in much detail here but I'll mention some basics. First, we need to determine the typical spectral sensitivity (visible and UV) of collodion emulsions used during the period. This is a straightforward matter. A much more complex problem is converting these images from monochromatic blue to panchromatic then concomitantly adjusting grey scale and gamma, etc. for accuracy.

As I see it, there are two levels to doing this. A quick short term approach that would significantly improve the appearance of the photographs and a longer more accurate one that would involve AI. Both methods would use the measured characteristic properties of the images (collodion spectral response etc.) and information that is known about objects in the photographs (these would act as known references). For example, we still have original Union soldiers' uniforms from the War, the dark blue colour of their jackets and the light blue of the trousers would provide reasonably precise colour references. The longer term approach would use AI learning in conjunction with image databases that consist of millions of objects whose colours and characteristics have been accurately documented. (I've little doubt that in the long term we'll be able to reconstruct full and essentially accurate colour images and films from their old monochromatic and panchromatic counterparts.)

The reason for why we would want to do this is obvious. If anyone wants an example I'll provide one.

† Over recent years I have spent what probably amounts to an inordinate amount time working with and trying to correct these old monochrome photographs in Photoshop, the GIMP and other image-processing software with the purpose of adjusting them to maximize their presentation (fixing gamma, optimizing black-crushing, reducing dust and scratches and such), and I can assure you that it's a very difficult job to correct them in ways that I consider would provide reasonably accurate results (i.e.: by today's high quality B&W standards). I'd have hoped that by now someone would have written a Photoshop plugin specifically for the purpose of processing blue-sensitive monochrome photographs but to my knowledge no one has done so (that this is the case is perhaps testament to the fact about how difficult it is to do if one has to maintain a reasonable degree of accuracy).

The nature of these old collodion emulsions is such that it is almost impossible to fix them by applying adjustments over the whole image. Instead, one reverts to correcting small 'local' areas within the image and adjusting individual objects one at a time (the high resolution of these images make these corrections easier although the extra detail makes it a much more tedious job). It goes without saying that when one processes images in such fine detail (which can take many hours) that one gets to know their contents in quite intimate detail—in all probably in much better detail than the photographer himself.

I recall such an instance when correcting a large 85MB TIF file of a somewhat mundane wet collodion photograph from 1865 of Union officers posing for a photograph on top of a small flight of tiered steps that led into their building. Each step was supported at both ends and in the middle (there being no back support to them). This meant that from the photographer's position one could see through the space between each step to the ground behind them. Whilst editing this area of under and behind the steps I accidentally stumbled across a collection of rat droppings together with what appears to be a discarded thin candle-like taper that was still alight (perhaps some smoker had thrown it there moments before). Thus, I claim the dubious honour of having found the oldest known photograph of rat shit ever recorded on film (I've challenged others in the past to beat this 157-year-old record but to date no one has taken up the challenge).

If anyone is so disposed to view the photograph, I'll provide details and references.

Whilst my discovery is a trite and somewhat amusing example, the more serious issue is that one would not easily detect these objects in a large contact print taken from its glass negative of typically 10"x 8" in size. Perhaps it may be possible to do so with a magnifying glass but one would have to be looking in exactly the right place and that spot is definitely not 'centre stage' of the photograph. What's relevant is that these historic photographs contain considerably more information than is superficially obvious and that we need new techniques to extract the details with a level of consistency/reliability. In this instance we've discovered unexpected proof of a hygiene problem with rats in Washington during the Civil War not to mention the casual way smokers (or others) discarded their burning wares. (Whilst not a serious issue here, a similar scenario elsewhere could see a building destroyed by fire, thus this information could, say, be used to help explain the causes of other fires that occurred during that era).

Edit: an afterthought. There are other images in LOC's Civil War collection that I'm aware of where important information cannot be gleaned at first glance. I discovered one startling example when I was attempting to correct a somewhat overexposed sky in a photograph of a railway yard. After doing so the image revealed a row of telegraph wires that spanned the complete width of the photograph (they were completely invisible before I made the alterations). Given the newness, relative rareness and importance of the telegraph back then, this new information gives the photograph a whole new look and complexion not to mention having a newly enhanced importance from a historical perspective.

If anyone wants information about that reference then I'll make it available.

Again, this is another testament to LOC's excellent scanning in that these telegraph wires were not crushed (white clipped beyond rescue), which is so often the case with such images.

Incidentally, the top and bottom default clipping points in Photoshop of 1% is a distinct hazard when doing such editing. Why Adobe set these clipping points to where they are is somewhat understandable but why they never bothered to warn users about potential data loss—in that they manifestly lead to a loss of picture information (detail)—is anyone's guess. Frankly, it amazes me that almost no one has bothered to comment on or complain about this major problem.

andsoitis2• 4 years ago

the experience of color cannot be quantified

dylan604• 4 years ago

Just ask Dorothy.

thenoblesunfish• 4 years ago

Do you mean this in the sense that all perceptual experiences seem somehow unsatisfying when quantified? That is, we can study the sensors, we have some ideas about how the signals from those sensors are post-processed in the brain, we can do experiments to determine what things people can (or do) differentiate or detect, and we can observe that have enough of a shared experience of the experience to be able to discuss and use the concepts in practical situations. However, for all that, somehow we have the feeling that we're not able to describe the "real thing".

wcoenen• 4 years ago

This is referred to as "qualia".

https://en.wikipedia.org/wiki/Qualia

andsoitis2• 4 years ago

experiences seem somehow unsatisfying when quantified

experience is subjective. how do you quantify something subjective?

thebeardisred• 4 years ago

I recently began going down a rabbit hole for an art project.

NARA has some interesting technical documents on rasterization of documents for archival purposes (https://www.archives.gov/preservation/technical/guidelines.h...) which I was reminded of by the "Golden Thread" target at the bottom of the image (https://www.imagescienceassociates.com/mm5/software/GoldenTh...).

ABA (Always be attributing): My partner works at a research library and has been involved with digital archivist for much of their career. Also, they illuminated me to both "Golden Thread" and FADGI (Federal Agencies Digital Guidelines Initiative) standards - https://www.digitizationguidelines.gov/guidelines/FADGI%20Fe...

igouy2• 4 years ago

iirc the experience of color changes when adjacent colors are changed — context matters.

kloch• 4 years ago

One reason even well designed camera systems struggle to reproduce accurate colors is the relatively wide bandpass filters for each color. This is especially true for modern digital sensors with green, cyan, and clear Bayer filters.

While inexpensive to implement and coincidentally well matched to narrow-gamut encoding like sRGB, it significantly mangles red/yellow/green colors.

Filters that approximate the CIE 1931 standard observer effective band pass curves would be better (and work well with wide gamut encodings like rec.2020)

Here’s some detail I posted in a previous thread: https://news.ycombinator.com/item?id=31107643#31111607

The sensitivity curves for Kodachrome are narrow and a good match for the CIE standard observer RGB effective bandpass curves (ignoring negative red values):

https://125px.com/docs/film/kodak/e55-2009_06.pdf

Keep in mind the response chart is log and what we are looking for is the -3db bandpass.

Humans being confounded by color photography

Related Stories