Basic Contents

Computer Vision ... (Published 1992)

Simulated Human Vision..... Ian Overington

Location: Eastbourne. UK
ianoverington@simulatedvision.co.uk ............ www.simulatedvision.co.uk

Chapter 6. Edge detection and local motion sensing in the presence of random noise.


It has been shown in earlier chapters that, under ideal (zero noise) conditions, it is possible to
achieve very high performance in the sensing of local fragmentary edge position, strength,
orientation and orthogonal motion. This is particularly true for a form of preprocessing based on 1st
differences of illuminance, as opposed to the more conventional 2nd differences (Chapters 4 and 5).
However, it has to be acknowledged that, in many practical situations, the images presented to an
image processing system are far from ideal. In particular they are often contaminated with random
noise. The question must therefore be posed "How robust are these high fidelity performance
characteristics in the presence of high random noise?". A first attempt to answer this question was a
comparison of VISIVE processing with a variety of classical edge detection algorithms, using one
particular test target and a goodness of performance measure developed by B.Ae. Hatfield from one
proposed by Wong. Algorithms compared included the Laplacian, Sobel, Kirsch, Frei-Chen, Prewit,
Roberts and compass gradient operators, plus 3 x 3 and 5 x 5 directionally averaged Sobel operators
(that is, groups of Sobels of similar orientation applied over 3 x 3 or 5 x 5 neighbourhoods). It was
shown in that study that 1st difference VISIVE processing was more robust against noise than any of
the other algorithms tested, whilst 2nd difference VISIVE processing was similar to many classical
operators. However, the performance measure used was itself open to some question. In
particular, it did not permit fully objective assessment of certain capabilities of VISIVE processing
which have been discussed in previous chapters and similar operators utilising extended receptive
fields. It was therefore considered necessary to carry out a controlled study of the effects of random
noise for VISIVE variants, along similar lines to the
basic performance studies reported in Chapters 4