2D Cellular Automata for Feature Detection
This is a Demonstration of the ability of 5-neighbor two-dimensional cellular automata to detect simple features. A rule and its rotated counterpart analyze the array at the left for horizontal and vertical characteristics. The panels at the right show the "activations" or percentage black at each evaluation step. The number to the right is the integral of the activation graph, or the combined activations of all evaluation steps.
This is a biologically inspired machine-vision scheme, designed to broadly mimic the human primary visual cortex using five-neighbor 2D cellular automata. The sample region (where is odd) is evaluated times. On each evaluation, the side length decreases by two, so the final point in the activation graph is a binary representation of whether the model "believes" the line to be horizontal or vertical, or whether the number of horizontal features has passed some threshold.