cubic in form. However, a cube is the most unsuitable
shape for shade sorting because:
(a) The colour difference between the corners and the
cube centre is 1.73 times larger than that between the
face centre and the cube centre
(b) The volume of a cube is comparatively small in
relation to its diameter, leading to an excessive number
of cutting lots [12,13].
In view of this, a sphere was used in the first microspace
shade sorting system. The colour populations were
grouped together to form spheres and then arranged in a
cubic close-packing form. Unfortunately, space between
the spheres was created and any specimens lying in this
space could not be sorted. Smart overcame this limitation
by putting the colour population into the polyhedral form
nearest to a sphere: rhombic dodecahedron [14].
An ideal block shape, having a maximum volume at a
given diameter plus the ability to be closely packed without
any intervening spaces, for shade sorting is known as a
truncated octahedron. It consists of six square and eight
hexagonal faces with each side having the same length. The
truncated octahedron microspace shade sorting programme
was used in all shade sorting programmes supplied
by Instrumental Colour Systems in 1984 [14]. Two
years later, the so-called CMC(2:1) microspace data was
used for the CCC shade sorting [15]. This system is
regarded as a grid-free shade sorting system as it relates the
absence of the predetermined boxes for separating coloured
samples. The population of colours is designed so that nearest
neighbours coalesce into clusters which then merge
with other clusters to produce larger clusters progressively.
Pairs of clusters with the least distance between each other
are successively chosen so that the resulting clusters possess
the least colour difference between all members of the same
cluster. Addition or removal of the underlying colour population
may cause a shift of colour clusters. The clustering
process continues until it exceeds the preset permissible
colour-difference tolerance. As a result, all clusters within
the permissible colour-difference tolerance merge together
and no more clustering proceeds when the combination
exceeds the permissible colour-difference tolerance.
Although the latest development in CCC shade sorting
provides an extremely effective method, i.e. a hierarchical
clustering algorithm for shade sorting, no systematic
study has been undertaken into setting up permissible
colour-difference tolerances, i.e. optimum tolerance limits.
Hence the present study was carried out.