Improved Resistive Fuse Grids
for Image Segmentation

Matthew Exon


The field of electronic vision is an area of intense research interest around the world. Usually this is seen as a high level task, to be undertaken by computer software. However, the achievable performance of these systems is limited by the intrinsically serial nature of computers. The potential exists to dramatically improve the performance and applicability of machine vision by implementing the processing algorithms as analogue circuits, hardwired in VLSI.

The purpose of this honours project is to investigate the implementation of neuromorphic vision chips that use analogue electronics to approximate digital visual processing algorithms. A design that would extend and improve the currently available designs was sought. The circuit is intended to be implemented on the same die as the photoreceptors that convert the light intensity to electronic signals. Thus, the entire chip could be used to provide a high-level description of the scene instead of a raw intensity image.

Using a relatively simple modification to existing network designs, a network was created that showed an ability to detect edges and regions in a manner much closer to the common sense decisions made by a human being.

Matthew Exon 2004-05-23