Professional Documents
Culture Documents
Multidimensional Mapping of Munsell Colors Varying in Hue and Chroma'
Multidimensional Mapping of Munsell Colors Varying in Hue and Chroma'
.93 for O, and .85 for M. Besides, the plots will be considered as lying in a real Euclidean
of £ij with Sji showed in all cases that the space of r dimensions if B* is shown to be of
points plotted were scattered along the line rank r and to have no large negative latent
passing through the origin with the slope root.
of one. There was a high degree of consist- Two B*s were constructed for Id and for
ency in each 5 to express 8,7 as either xmo K, O, and M. In accordance with Equation
or #,•,()). The criterion tells us nothing, 7, B* was factored by the centroid method in
however, concerning whether the relation- the case of Id and by the multiple-group
ship between x and S is really as simple as method (Harman, 1954) in the other case.
given in Equation 1. Even if 5y is replaced Three factors were extracted in both cases.
by f(Sii), a function of &u, the whole pro- For the following reasons, however, the third
cedure of the scaling still holds and, instead factor could be discarded and we could con-
of Equation 6, f(Sa)e will be obtained as £«. clude that B* was of rank two, except for
In this case also the criterion will be met variable errors, for Id as well as for K, O,
since f($ij) is equal to f(Sji). It is a matter and M: (a) The second factor residuals all
of assumption that the observable x is put seemed sufficiently small; (6) The contribu-
in Equation 1 as proportional to the latent tion of the additional factor was relatively
variable 8. No empirical verification was small. The ratio S,'a2i3/Zi<z2,-i was .13 for Id
possible in making this assumption. and .17 for K, O, and M. (c) The criterion
The multidimensional scaling was applied for B* to be of rank two is given by Torgerson
to the data of Id and also to the pooled data as follows:
of K, O, and M. The rf«'s of the three 5s
were pooled as follows. In order to reconcile 2 S (&*«)« = ( S »»,-!)» + ( S atitf
the difference in the Unit e, the d./'s of each 5 i t i i
were plotted against the d,-/'s of Id and the + 2( S aao.ti)2 [8]
i
former were multiplied by a constant so that
the points plotted were scattered along the In the case of Id, 81.18 X 10' was obtained
line passing through the origin with the slope for the left-hand term and 76.70 X 10' for
of one. By taking the unit of the scale of the right. The coincidence seems close
Id as reference, the scales of all 5s were thus enough to discard the third factor, though
defined by a common standard. The con- it is a little worse than the coincidence in
verted dij's of the three 5s were averaged Torgerson's experiment (1958). The ratio
to give the Common Distance Matrix D. between the left- and the right-hand members
Multidimensional scaling.—Let i and k be of the equation was 1 to .944 in the former
alternative subscripts for stimuli (i, j =* 1, and 1 to .981 in the latter. For the data of
2 n) and let &,-,- be the scalar product K, O, and M, the criterion would be inappro-
of the vectors from an origin to Points i and j priate because of the method of factoring.
in a real Euclidean space of r dimensions. In order to apply the criterion, it is desirable
As suggested by Torgerson (1952, 1958), it that the a{p's in the equation be the results
is convenient to place the origin at the cen- of the method of principal axes or, at least,
troid of all of the stimuli. Let the scalar of the centroid method, (d) No particular
product thus defined be denoted by &*,-,•, relationship was found between an and the
then &*<,• can be derived from Matrix D by fluctuation of the stimuli in Munsell value.
the formulae given by Torgerson. Once
Matrix B*, whose elements are &*,,-, is ob-
tained, it becomes possible to determine the RESULTS AND DISCUSSION
dimensionality of the space and to obtain
Oif, the projections of points on an arbitrary The configurations given by .4s are
set of orthogonal coordinate axes, because B* shown in Fig. 2 and Fig. 3 where the
can be written as follows: coordinate axes are omitted. Instead,
B* = AA' [7]
radius vectors are drawn so as to run
through colors of the same hue and
where A = (aip) and p = 1, 2 r. A' a series of curves is also formed for
is the transposition of A. It is important chroma. Since the scale units have
to note that Equation 7 does not hold unless been equated as stated before, the
B* is positive semidefinite, that means, r
latent roots of B* have relatively large posi-
two configurations can be directly
tive values and the remaining roots vanish compared with each other.
except for errors. In other words, the colors It is interesting to observe that
326 TAROW INDOW AND TSUKIKO UCHIZONO