Algorithmic Typewriter Art: Can 1000 Words Paint a Picture?

Jules Kuehn, David Mould

Keywords: Non-photorealistic rendering, ASCII art, typewriter art, optimization, NPR

TL;DR: We convert images into typewriter art that can be physically reproduced in four overlapping, staggered layers of type by having the typist move the paper.

Abstract:

We develop an optimization-based algorithm for converting input photographs into typewriter art. Taking advantage of the typist’s ability to move the paper in the typewriter, the optimization algorithm selects characters for four overlapping, staggered layers of type. By typing the characters as instructed, the typist can reproduce the image on the typewriter.
Compared to text-mode ASCII art, allowing characters to overlap greatly increases tonal range and spatial resolution, at the expense of exponentially increasing the search space. We use a simulated annealing search to find an approximate solution in this high-dimensional search space. Considering only one dimension at a time, we measure the effect of changing a single character in the simulated typed result, repeatedly iterating over all the characters composing the image.
Both simulated and physical typed results have a high degree of detail, while still being clearly recognizable as type art. The accuracy of the physical typed result is largely limited by human error and the mechanics of the typewriter.

Confirm No Double Submission: Yes

Withdraw On Reject: Yes