We have produced an experimental implementation of a massively-parallel reaction-diffusion processor which performs one of the most essential parts of shape recognition - construction of a skeleton. A skeleton is a unique, stable and invariant representation of a shape, therefore computation of the skeleton is an essential tool of computer vision. Skeleton computation is a typical 'natural' spatial problem that can be solved with the use of biological, chemical or physical phenomena. One possible approach - a reaction-diffusion based computation - is explored in this Letter. A contour is represented by a concentration profile of one reagent, a planar substrate is mixed with another reagent. The reagent, representing the original contour diffuses to form a coloured phase in a reaction with the substrate-reagent. However, at sites where two diffusion wave fronts meet no coloured phase is formed and the substrate retains its uncoloured state. These loci of the computation space represent a skeleton of the given contour. In the Letter we only describe a laboratory prototype of a reaction-diffusion processor that computes a skeleton, no further tasks of image processing are undertaken, one could say we have designed an unconventional chemical pre-processor for shape recognition. © 2002 Elsevier Science B.V. All rights reserved.
Adamatzky, A., de Lacy Costello, B., & Ratcliffe, N. M. (2002). Experimental reaction-diffusion pre-processor for shape recognition. Physics Letters A, 297(5-6), 344-352. https://doi.org/10.1016/S0375-9601%2802%2900289-X