Lossy Description and the Hermeneutics of Images

February 15, 2015

When taking a photo with your phone, it automatically submits the raw data captured by the lens to a regime of processing. Most raw data is lost, as storing the complete amount of information that the camera could capture with every photo would take immense amount of storage space. The popular JPG format, for instance, increases contrast and exposure, reduces noise and sharpens the image while compressing the overall file size. While the resulting picture is more palatable to contemporary aesthetics, the output is merely one interpretation of the full spectrum of data that the camera could have presented, had the image been processed differently. We should not forget the importance of compression here — and its lossy quality — but to this we will return later.

Services, such as Instagram, take this image processing further, applying other filters to the JPG data and highlighting the information remaining in the photo in distinct ways. It has not been lost on many commentators that Instagram’s particular style of filters seem to capture the nostalgic effects of photographic processing techniques prior to the digital era. This aura (to evoke the Benjaminian term) has resonance, not simply for its ability to improve upon the legibility of the image for human consumption (as much of the JPG processing is focused upon) but also for its capacity to recall certain aesthetic regimes.

That these regimes are themselves the result of technical progress in the development of photographic processing, particularly related to the film era, and its different techniques of reproduction — such as the polaroid image — is itself interesting. For, the material artifacts of these technical regimes (the modes of their processing) are themselves the result of particular technical descriptive models. That is, the problem of how to capture light on a surface becomes decomposable to a particular mechanism, which can be described by a model that can be implemented in practice, and resulting in the artifact that captures — in one particular way (of a range of potential solutions ) — light on this surface.

However, as successive technological movements improve and add to the potential means of managing this process, the variance in technical capacity becomes associated with the genealogical dimensions of the culture in which it is immersed. The historical indexing of certain technical functions with the aesthetic of image capture become a part of a normative understanding of the historical era and culture of which they are a part. We might peg this indexing with the functional insufficiency of existing techniques to capture data that is revealed by the later developments of imaging technologies. Most obviously, consider the turn from black and white era photography to color photography, and how the perfection of the mechanisms to capture color in photography became widely commercially available in conjunction with the 60’s, the space race, and modern media culture.

It then is the lossy quality — what is not captured in the image — that comes, in part, to define its association with certain eras. What is peculiar about the Instagram image is its capacity to choose amongst these lossy qualities – it is capable of switching between various modalities of image loss to pick and choose a mode of reproduction that is its reference to a unique technical historical development. In this then, it seems to operate on a meta-level. The processing itself, however, is hidden beneath the sheen of the interface. An immediate connection with the mechanisms of its production, unlike in older techniques, is lost. Reactionaries, who wish to constrain technical production to the empirical interface of the human-scale may bemoan this transformation, but this is not my point. The descriptive mechanisms for these operations still remain, if opaque to most casual users of this technology.

The transformation from the technical mechanism to its aesthetic is itself the transmission of a new type of information. The new information is the conscious reference to the previously existing historical and cultural conditions as an conceptual-aesthetic gloss on the image. While the technical production of a particular historical form can be described by the opaque mechanisms of its production (that is, for digital images, the algorithmic processing of the filters that result in the look of the image) the normative content of the image thus processed cannot be reduced to such mechanisms. This is because the normative cultural-historical indexing of technical-aesthetic regimes comes freighted with conceptual content that is not described by the technical-causal models of production.

This is not to say that it does not exist, or could not be understood as a part of the conceptual dimensions of the artwork, but access to these conceptual dimensions does not derive inherently from the technical description of the artwork. Rather, it plays upon the normative conception of the aesthetic presentation as a cultural-historical index. This kind of conceptual understanding of the work relies upon the production of the aesthetic as itself a kind of language — that is, a film shot with a modern digital camera could be filtered to present in black in white, as a conscious attempt to reference earlier eras of film-making. The technical means of this production, and a mechanistic explanation of how this effect is achieved, tells me very little about the intention of the filmmaker’s conceptual process in making this decision.

Consider new developments in image processing: The semantic descriptive indexing of pictures by machines. This is a developing technology that has proven extremely difficult to implement, for it requires the structuring of multiple hierarchical models of image processing sequenced in relation to one another AND linguistic models. This image processing must be capable of distinguishing figure from ground, recognize patterns, and describe objects and regions in the image data. While JPG compression highlights general qualities of the image (such as its contrast and exposure) at the expense of other qualities, the processing of the image into a linguistic description must become lossy in a new sense. Given an image of a man playing fetch with a dog in park, the algorithm must be able to pick out “dog”, “man”, and “park”, generalizing from the countless other potential features that might be descriptively derivable from such an image.

The selective highlighting, however, of the image into the framework of language, while lossy in terms of the breadth of potential information available in the image itself, provokes new conceptions of the image, derived from these generalizations ramified through language. Once we are in the space of the inferential nexus of language, we cannot but help to draw inferences with the manifold of conceptions we have in reserve. The production of artworks comes from the sapient capacities of language users, not the mechanical reproduction of machines (this is not to say that machines could not some day produce artworks, or that we cannot derive aesthetic satisfaction from the production of machines). Artworks are produced, not solely from their material properties, but in reference to conceptual discourses, whose normative dimensions cannot be shorn from their understanding.

The attempt, therefore, to reduce the reading of aesthetic images to a mechanistic description of the production of their surfaces, in purely causal descriptive language, is an impoverished and eliminative prescription for the production of art. In attempting to maintain a scientific accuracy of description, it loses out on the various conceptual dimensions whereby Art accesses distinct normative hierarchies of description.