Basic Contents

Computer Vision ... (Published 1992)

Simulated Human Vision..... Ian Overington

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

Chapter 8. Image Preprocessing


It is frequently the case that images which are to be offered to a computer vision system are already
prerecorded (e.g. recorded on video tape from thermal imagers, standard commercial video
equipment, photographs etc.). This is particularly true of virtually all real world images which are
used for
research studies into computer vision. These input images are therefore presampled at
discrete spatial and temporal intervals,
may include patterning due to photographic grain,
photodetector noise etc., and in general, incorporate a (usually unknown) small amount of image
blur. Now display / observer interface studies for
human visual observers have shown a number of
important facts regarding perception of discretely sampled images as follows.

A very widely used form of modern image recording and replay is by raster scanning (essentially as
used in commercial television). As presented to the observer, such raster scanning will provide an
image which is characterised by a modulation structure across the raster lines and an angular
subtense of the raster line pitch. The former of these is determined by the internal setting up of the
display, whilst the latter is governed by the display size and the observer viewing distance. It must
be anticipated that both these factors may have some effect on observer performance. To our
knowledge no major studies have been carried out on the effects of detail across the line structure,
although the effective modulation depth, which is related to the sharpness of the scanning spot, has
been shown to have large
subjective effects. What has been researched more objectively is the
effect of raster line subtense for a 'typical' form of raster structure having a 1:1 mark/space and a
modulation depth of approximately 0.8. When human observers view raster scanned images their
thresholds of detection and recognition are very dependent on the relationship between the scale of
raster line pitch and the foveal receptor matrix (which is the human visual system's spatial