Abstract

This paper describes a new charge-coupled device (CCD) focal-plane image-processing technique that can carry out the convolution of an image with two-dimensional Gaussian kernel functions. Such Gaussian convolution operations are the computationally intensive steps in the difference-of-Gaussian procedure employed in artificial intelligence algorithms for the detection and location of the edges of objects in an image. Devices using our technique can produce convolutions for a wide range of Gaussian widths over an arbitrarily large field of view in a time less than that required to detect the original image.

© 1985 Optical Society of America