Digital images generated with custom convolution code (available in browser at https://v21.io/r_g_b.html). February 2022.

This artwork is an algorithm that explores what I call "iterative convolution" — the process of repeatedly feeding the output of a convolution filter back into that convolution filter. This process is controlled by both myself and the machine: the fine parameters of the algorithm slowly shift by themselves, while I respond, controlling in real time the sharpness and scale of the transformation.

Convolution is one of the building blocks of machine learning, used in almost all AI processes involving images. One could say that much of machine learning is just about trying to find the correct set of convolution parameters to achieve a certain task. Going further back, convolution is the fundamental technique underlying the blurring, sharpening and edge detection filters that will be familiar to anyone using image editing software. This r_g_b project is part of a series of works exploring the aesthetic implications of these image kernels. What forms do these processes create; what does it look like to wander randomly through this state space?

The name ‘r_g_b’ is derived from the seed used for these images: just three dots, one of each of red, green and blue. Each colour channel is processed in parallel, the forms varying independently, but also evolving according to the same parameters. This builds on previous work I have made with a single greyscale channel being manipulated, and also extends the system to function across larger scale forms, bringing out, in places, an almost analogue graininess from a purely digital system.