Vague Terrain

[detail of p. 8 from standard test pages for early fax transmission]

Our common processes and protocols for digitization rely on a finite corpus of sample images, movies, objects and texts. Before consumers had access to such common formats as JPEG image compression or Quicktime videos, researchers stared at a few choice test subjects for hours and hours, until these images became mere indicators of fidelity, devoid of content.

For new viewers, these dead images, more index than symbol, are as arbitrary and puzzling as any meme. Like image macros, they point more to previous viewings than to the strange referents they contain. Yet the referents hang around like ghosts, and make me wonder how our formats would differ if other images had happened to be lying around the labs.

If you hold to the blunt criticism that digital life carries more information than context, then these test images are the ur-symbols of our data-cult. I find it pleasurable and useful to reach back and look at these objects out of their original research setting, to let questions of content and context rush back in.

If lena.tif is the most famous of these test subjects, other common test subjects in image compression reveal how poorly-suited the "Lena" TIFF is to demonstrating real-world fidelity problems. Right up there with lena.tif in popularity, if not in infamy, are these three images. Like Lena, each are typically provided at 512x512px, 72dpi, in uncompressed TIFF format. Compared to Lena, they present clearer technical problems to solve.

[from left: baboon.tif, goldhill.tif, peppers.tif]

Looking at these three - often found in the same directory online - we see both a range of technical challenges and a range of hoped-for applications for digital imaging.

Baboon.tif (actually a mandrill) and peppers.tif offer some obvious opportunity for color correction and testing through contrast, though it could be asked - how would most people know exactly how blue the mandrill should be? In fact, those who set standards for compression formats rely on more than subjective judgment of an individual image. Standards bodies such as the Joint Photographic Experts Group have released other test images that contain embedded data, hidden characters and markers to be sent and received through the encoding process, as in steganography.

Baboon.tif brings some texture challenges, and goldhill.tif (a photograph of a street in Dorset) some opportunities for comparing the same algorithm on plants and architecture. There's also some good opportunity in goldhill.tif for establishing clear aerial perspective and depth of field. Peppers.tif provides us with our only reflective surfaces in the early sets.

So a set of desires from imagery emerges here, both in form and content. What might we want from digital images? Apparently we would want, in addition to faithful rendering of exposed human skin, easy depiction of clearly marked and identifiable color, the rendering of deep space and diverse surface textures, glosses. We might also see here some imagined future uses for digital photography - education(baboon), tourism(goldhill), and sales(peppers).

Others images pop up from time to time, and some companies work from their own batch of recognizable and abstract imagers. Other common characters include the Cameraman, a fishing boat, and someone called Zelda.

[from left: cameraman.tif, boat.tif, zelda.tif]

If we look back a step in technological history, we find the eight standard test images [PDF link] for fax technology, which also provide both a range of technical problems and a revealing survey of hoped-for applications.

[from left: CCIT p.8, CCIT p.7, CCIT p.2]

Looking at these fax subjects, we're also reminded of the origins of image compression in communication theory - and how all our visible digital stimuli have roots in the encoding and transmission of audio data. Just as in the longer media histories of McLuhan and Ong, the ear precedes the eye in the story of digitization.

Today we may throw around the metaphor of image as information with abandon, but the language is not figurative in these cases. In this purest form of the analogy, we see how the understanding of images as information can literally drain the subject away. When I imagine these images proliferating through labs, machines, academic papers, and conference screens, I'm struck by how freely they fly, wholly unencumbered by intellectual property issues, wrung free of aura or content, the purest form of visual data. Yet assembled again into a collection, their mysterious content and context rushes back, through the pleasure of the wunderkammer, and the arbitrary nature of science's subjects.

In the next installment of the series, we'll look at the test subjects for 3D modeling - and learn how many ways one can slice a Buddha. Eventually we'll look at examples from video and text as well, before moving to the recent return of proprietary approaches to test images, in facial recognition and data-mining.

Additional installments of this series: Part One, Part Three.