Chapter 16. Differential Texture Analysis.
It is generally agreed that the different textural properties of parts of a scene are an important cue to
the underlying scene content. Such properties, which include both local statistical fluctuations and
more global structural elements, may be used for purposes of scene segmentation, recognition and
interpretation. In particular, the local statistical fluctuations are crucially important for camouflage
purposes. In this chapter we shall mainly be concerned with the sensing of such local statistical
fluctuations and with locating changes in local statistical means which may signal region boundaries.
It specifically excludes considerations of structural analysis of texture, which has been deeply
studied and reported elsewhere. Also it makes no attempt to synthesise or simulate texture, topics
which have again been addressed by others.
Most conventional methods of sensing and analysing local texture start with a square matrix
sampled image, frequently coarsely sampled, and seek to quantify the textural properties in terms of
first and second order statistics of individual pixel energy levels. Conversely, over many years our
studies into what are the important factors controlling human thresholds of visual performance (as
discussed in Chapter 2) have led to an acute awareness that the important scene information used by
human vision, both for local texture perception and for form recognition, is related to fragments of
local maxima of illumination differences in a hexagonally sampled image, this image being