In Reconstructing Vectorised Photographic Images¹ three computer graphics researchers describe a method for representing moving images by encoding their isochromatic contours as vectors.

Illustration from: J. W. Patterson, C. D. Taylor and P. J. Willis, "Reconstructing Vectorised Photographic Images"¹

This paper focuses on the problem of rendering from the isochromatic contour form of a vectorised image and demonstrates a new fill algorithm which could also be used in drawing generally. The fill method is described in terms of level set diffusion equations for clarity.¹

Even if this research seems on hold at the moment, it's quite an unusual ―yet fantastic― loop-back that is set out here. Video devices measure and capture light signals, samples taken regularly on a grid. Important for our context, vectors are decribed and recorded mathematically, in a number form that is not so different to light signal mathematics. Compressing the large quantities of data that the photosensitive grid sampling method produces has always been a challenge, not to mention the uphill battle that these Codecs have to deal with as the resolution—and frame rates— of captured images keeps growing.

We also note that machine learning is now in use to augment in large proportion (up to 4×4!) resolutions by guessing what information is missing from the deduction database it has built.²

At the reverse, the idea to limit shapes and interstitial elements as a way to simplify an image could seem interesting or in the contrary frightening. Moreover, the decoding portion of this new method suggests in a tempting way that we can reproduce a vision from mathematical lines. Once ordered and combined, this gets us probably far from to the way living brains works but maybe brings us in an interesting new zone for the production of images.

Let's dream of a camera, a metasensor, which would be vectorial as raw output.