Machine Vision Without Pixels

Tyler C. Folsom

Abstract: Hubel and Wiesel received the Nobel prize for describing how individual cells in the cat brain respond to light. Other researchers have derived mathematical models that describe brain cells as linear filters. My work investigates how these filters could be used in machine vision. The algorithm involves covering an image with an overlapping circular grid. Computations are performed on each circle (NOT at every pixel). At every circular receptive field, the subimage is convolved with steerable oriented filters that approximate the first and second derivatives of a Gaussian. Assuming that the subimage approximates a bar or an edge, it is possible to find its orientation and subpixel position. When we examine features normal to the dominant orientation, we can locate corners or interest points. The main output from the algorithm is a list of feature points, but the software also produces an image for quick verification. Preliminary investigations indicate that its speed is at least competitive with the Canny edge detector and Harris corner detector.

Brief tutorial

Longer explanation

Download code and documentation






Last update: August 4, 2003