Scan transcription of two-dimensional shapes as an alternative neuromorphic concept

TL;DR

This paper proposes an alternative neuromorphic approach for shape recognition that moves beyond traditional orientation-based contour filtering, aiming to inspire new encoding methods for shape information in neural models.

Contribution

It introduces a novel concept for encoding shape information that challenges existing orientation-selective neural network models in neuromorphic computing.

Findings

Proposes a new shape encoding method for neuromorphic models

Challenges traditional orientation-based contour filtering

Offers a biologically plausible alternative for shape recognition

Abstract

Selfridge, along with Sutherland and Marr provided some of the earliest proposals for how to program computers to recognize shapes. Their emphasis on filtering for contour features, especially the orientation of boundary segments, was reinforced by the Nobel Prize winning work of Hubel & Wiesel who discovered that neurons in primary visual cortex selectively respond as a function of contour orientation. Countless investigators and theorists have continued to build on this approach. These models are often described as neuromorphic, which implies that the computational methods are based on biologically plausible principles. Recent work from the present lab has challenged the emphasis on orientation selectivity and the use of neural network principles. The goal of the present report is not to relitigate those issues, but to provide an alternative concept for encoding of shape information that may be useful to neuromorphic modelers.