Fig. 3 shows this kind of situation, where the histograms
have two or three modes. In both cases the
images cannot be thresholded globally, irrespective of
selected global bilevel thresholding algorithms. These
images are classified as class 2, and a global thresholding
algorithm cannot be applied. Instead, a thresholding
algorithm development has to be based on local
information (e.g. adaptive thresholding). Until now
there has been no thresholding (speci®cally adapted to
image analysis of aggregates, Wang, 1997), which
could process a mix of these two types of images simultaneously.
We developed a special algorithm for this
application. In what follows, we ®rst brie¯y describe
how to choose an existing global thresholding algorithm
that is applicable to the ®rst class of aggregate
images; secondly, we describe how to threshold the second
class of aggregate images based on the selected
global algorithm, where we use information about the
size, shape and range of gray levels of the particles. In
addition, we compare our thresholding algorithm both
to an adaptive thresholding algorithm and to another
standard thresholding algorithm, Gonzalez and Wintz
(1987), based on boundary characteristics. Finally, we
present ®eld test results, discussions and conclusions.