The vast majority of the images we encounter each day are natively neither physical nor even visual, but are composed of various forms of hidden textual data. This has a wide range of implications, not least for the development and understanding of computer vision systems.
In collaboration with creative coder and generative artist Matt DesLauriers, we are developing a system which transforms recognisable photographs into new visual forms. These outputs appear highly abstract and at times even random, but contain essentially the same data as the source image, only re-rendered into new forms, largely unrecognisable and unintelligible to human visual cognition.
The outputs from this system are still, in a very literal sense, photographs, but in rendering them in this way we hope that human viewers will start to appreciate some of the ways that computer vision is very different from our own. In particular the way that these systems see images not as a series of symbolically significant visual elements adding up to sum greater than their parts, but as an impersonal aggregation of tonal, chromatic and spatial information.