Of the readings this week, I would like to focus specifically on the Diana Kamin article: “Mid-Century Visions, Programmed Affinities: The Enduring Challenges of Image Classification” and compare her insights with a recent example of a large image classification system which has a direct through-line to another broad and pervasive category of photography. I will also highlight my personal user experiences in the exploration of the two online archives described in Nina Lager Vestberg’s article “Ordering, Searching, Finding” as well as the British Library’s Endangered Archive Programme (referenced in Allison Meier’s post “Four Million Images from the World’s Endangered Archives”.)
Before getting into those topics, I would like to quickly comment on the other readings for this week starting with the John Tagg paper, “The Archiving Machine; or, The Camera and the Filing Cabinet.” I did not care for this article! I found his writing to be overly dense and dry and hard to get through. This writing is successful in denoting the major innovation achieved by “the modern vertical file” and its associated cabinet and I don’t wish to trivialize that by any means. However I found the self-seriousness of this writing to be formidable and the conclusion (“one might say that the archive must and must not be the horizon of our future”) to be dissatisfying. (Tagg, 34) After reading it a second time I realized that there is a video on Vimeo of the author delivering this paper word for word which makes it all slightly more digestible but not less annoying. Apologies for this bit of editorial!
However I appreciated the in depth background on Alphonse Bertillon, who served as the director of the identification bureau of the Paris police at the end of the 19th century — his use-case described the impossible task of trying to sort through a collection of over 100,000 photos owned by the police to find a single criminal highlighted the need for and radical innovation of the classification and organizational structure of a “Bertillon” cabinet. Tagg convinces of the massive impact made by the debut of the modern vertical file at the Chicago World’s Fair in 1892 and the broad changes to systems of organization that followed. This emphasis on the need for “a systematic order” for storage/retrieval and classification specifically for collections of photographs is a major theme through this piece and the rest of the readings for this week.
Tagg also briefly lingers on the word and concept of “capture” in terms of systematic/organizational apparti (apparatuses?) but I followed through to its function in photography specifically — each photograph is a captured moment. Vestberg expands on this in her piece by noting that “every photograph is … also a mini-archive within the archive” (476) which, when a photo is housed in a large systematized, organized collection of other photographs, strikes me as very true. The process of “capture” in terms of photography and artistic process versus pure documentation is described in Douglas Crimp’s “The Museum’s Old, the Library’s New Subject” when comparing the works of Picasso and Ansel Adams in the context of the Museum of Modern Art’s fiftieth anniversary. The conversation of “taking a picture” versus “making a picture” (Crimp, 71) inspired me to contemplate when the photograph becomes a data element versus a constructed artwork, or when it can serve as both. This is illustrated by the anecdote that Crimp describes of coming across Ed Ruscha’s photography artbook Twentysix Gasoline Stations in the transportation section of the NYPL. Crimp concludes that “the fact is there is nowhere for [this book] within the present system of classification” (78) which is answered by Anna Sophie Springer who follows from Crimp in her article “Original Sun Pictures: Institutionalization” by stating that Twentysix Gasoline Stations should probably be shelved in the Photography section of the library, as “photography has advanced into a proper, canonical genre.” (Springer, 123)
Moving on to the Diana Kamin article. She successfully illustrates right out of the gate how we are currently drowning in an unorganized mass of digital imagery — “In the year 2015 alone, more photographs were taken than in the history of analog photography combined” and noting that “1.8 billion pictures are uploaded to social media (our contemporary universal image archive) every day” (Kamin, 311) — and the ongoing challenge to classify and categorize photographic imagery. She describes this problem by surveying two approaches to image organization undertaken by two librarians in the mid-20th century. We then meet Bernard Karpel, a librarian from MoMA, and Romana Javitz, who headed the Picture Collection at the NYPL.
Without re-hashing and summarizing every point of the article here, Kamin categorizes Karpel’s proposed system of image classification as a model based in the “discourse of affinities” whereas Javitz’s user-centric approach as housed within the “discourse of the document.” To briefly re-cap:
Karpel’s insane proposal for a visual system of classification that he hoped would become as ubiquitous as the Dewey Decimal system was based on “an ‘objective’ eye that sees [the visual object] without the clouding of context to evaluate only form.” Kamin notes that this perspective was probably bolstered and built up by Karpel’s “long career at MoMA, where aesthetic formalism was the dominant methodological approach for analyzing works of art.” (Practically speaking, this proposed visual system sounds like a complete mess to me — which Javitz rightly argued would discount a broad section of the general public who did not have the specialized knowledge that would be required to communicate inquiries into any collection of objects. Karpel proposed a basis of “aesthetic evaluatives (qualities such as tactility, transparency, and multiplicity) [that] would be arranged along the axes of his primary cataloguing tools: ‘affinity’, ‘polarity’, and ‘sequence’, which would address, respectively, formal consonance, dissonance, or relationality between two or more pictures.” (314) Understandably, this very complicated system involved duplicate cataloging cards and is totally impenetrable for a common user.
Javitz’s method for image classification was just the opposite. Kamin describes how she was vehemently opposed to hierarchical structures of organization/classification, as illustrated in the description of Javitz avoiding subheadings as much as possible: whereas previously “‘Lakes would be found under ‘F’ – Forms of Land and Water – Lakes’”’, Javitz deemed that Lakes should appear under ‘L.’
It is not hard to understand how this common sense and user-focused approach to visual classification of images won out over the very tricky visual basis of Karpel’s system of affinities. Javitz advocated that photographs be considered as documents first and classified as such. However there remains a lot of space between these two arguments in our current moment with technology. When trying to parse the what’s surely by now 2B + images that are being uploaded every day, machine-abled automatic indexing of images is a problem that many are working to solve. Javitz’s and Karpel’s approaches are both being applied. Kamin notes: “Javitz’s discourse of the document is most frequently encountered in the keywording dominant on the internet (a system in which any image can be tagged with multiple identifiers, or in natural language, by its uploader), while the discourse of affinities is manifest in discussions around pattern recognition and machine vision.” (329)
Both of these approaches are visible when looking at project by Google that I have found myself thinking a lot about during the course of this semester. The Google Image Labeler was an interesting space on the internet while it was initially active between 2006 and 2011, and it has returned since 2016 in a new form that remains relevant in the conversation of indexing and classifying images.
In its original form, the Image Labeler was set up as a game between two users who would be automatically partnered by the software. A time limit would be set and then a series of images would appear on the screen. Each ‘player’ would then submit text describing the contents of the image. Points would accrue when the two players would agree on terms. There were no prizes! However this was an active space online, and it was effective in improving Google’s image search function. This was the company’s clever way to ‘gameify’ the very manual approach to keywording and indexing images that would be required in order to achieve robust search.
The Image Labeler was shut down between 2011 — among other issues there became pervasive abuse between players, spamming the game with words like ‘abrasives, entrepreneurialism, and forbearance, among others — and reemerged in a different form in 2016. To this day you can log-in to the Labeler and lend a human-based approach to visual indexing. The images that confront users today are confusing and not straightforward to answer. (How do I see fog in a photograph?) Machine learning and approaches based on similar visual forms, as Keplar saw and envisioned the world, has come a long way in the intervening years but the current Labeler gives some insight into what types of images the machines are still struggling to classify, and the categories of images that users are most interested in.
Image classification via keywording remains an important and valuable characteristic in the broad world of stock photography. Stock photo collections are larger and more prevalent than ever. These images surround us in the world, and the measure of success for the end user (usually commercial) depends on choosing the right one. In “Re-use Value” from Cabinet Magazine, Jenny Tobias highlights the importance of keywords in the stock image marketplace. Referring to the image at the top of the article, Tobias writes:
According to the original caption, it is also a Portrait of Otto Bettmann—About 2 Years of Age, With an Umbrella.
Little Otto in his skirts could not have known that he would start collecting images as a teenager, earn a Ph.D. at twenty-five, begin compiling a picture history of civilization while curator of rare books in the Prussian State Art Library in Berlin, flee Nazism to the United States in 1935, and shortly thereafter found the Bettmann Archive, a major purveyor of stock photography. Nor did he know that sixty years after its founding, the Archive would be acquired by Microsoft’s Bill Gates for his Corbis image archive, shipped from lower Manhattan to a climate-controlled Pennsylvania mountain for preservation and digitization, and then redistributed over the Internet, where little Otto can be found today, exactly a century after the photographer snapped the shutter.
The article notes that the only transaction on this photo of Otto was for personal use, so not commercial — but the keywords listed here go very deep beyond what is shown in the actual image, and such is the nature of keywording stock images.
As Vestberg writes in her “Ordering, Searching, Finding” about the user experience of navigating the iconographic digital databases of the Warburg and Conway Libraries, keywording is key for researchers and stock hunters alike — and there are challenges: “Since the keywording system for digital files closely follows the iconographic categories of the analogue files, there are a number of elements in any image that may not be included in the metadata because they have not, for whatever reason, been considered iconographically significant in the process of keywording.” (Vestberg, 477).
In the case of the stock photo of Otto Bettman, it seems impossible to be able to feed this image to a computer and have the machine know on its own to tag the image with “Prominent persons.” The central difficulty here is summed up by Vestbeg neatly: “accounting for what a picture shows is never the same as describing what it depicts.” (478)
I was inspired by this last article and its elucidated challenges with searching these archives for images of “arrows” in order to find representations of “Saint Sebastian,” and vice versa. I decided to explore the Warburg, Conway, and the Endangered Archive Programme (just for fun) with the search term (keyword) “Magic” and compare results. (I chose Magic because I love the Warburg system of classification a lot, particularly the subsection of “Magic and Science.”) My findings were varied. The Warburg iconographic database returned 696 results that contained ‘magic’ in the metadata, but there were a lot of duplicate results. The Conway Library surprisingly only had 23 results, and the EAP had just one higher than Warburg with 697 (also containing duplicates.) I have prepared the first ~ 20 results from each of these for you to see, here. I find the visual artifacts from Warburg to be the most aesthetically satisfying, though this journey of searching and finding through the varied organizational structures of these photo collections was more meaningful than the destination of visual results.
John Tagg, “The Archiving Machine; or, The Camera and the Filing Cabinet,” Grey Room 47 (Spring 2012): 24-37.
Douglas Crimp, “The Museum’s Old, The Library’s New Subject” in On the Museum’s Ruins (Cambridge, MA: MIT Press, 1993): 66-83.
Anna Sophie Springer, “Original Sun Pictures: Institutionalization” in Fantasies of the Library, eds. Anna-Sophie Springer & Etienne Turpin (Berlin: K. Verlag, 2014): 119-31
Diana Kamin, “Mid-Century Visions, Programmed Affinities: The Enduring Challenges of Image Classification,” Journal of Visual Culture 16:3 (2017): 310-36.
Nina Lager Vestberg, “Ordering, Searching, Finding,” Journal of Visual Culture 12:3 (2013): 472-89.
Allison Meier, “Four Million Images from the World’s Endangered Archives,” Hyperallergic (February 23, 2015).
Wikipedia contributors. (2018, August 7). Google Image Labeler. In Wikipedia, The Free Encyclopedia. Retrieved 20:12, November 20, 2018, from https://en.wikipedia.org/w/index.php?title=Google_Image_Labeler&oldid=853941020