The first step is to segment a beef rib-eye image
into lean and fat regions so that the feature
quantities of beef marbling can be extracted with
high accuracy. Kurosawa and Nakanishi (1995)
and Shiranita et al. (1996, 1997) were using fuzzy
inference and neural network, respectively, to
construct the algorithm for segmenting the beef
rib-eye image into lean and fat regions. Both of
these methods, in our opinion, can be dicult to
translate into simple algorithms. Moreover, both
fuzzy inference or neural network assume several
uncertain elements such as the initial value, a local
solution, etc. Because of this, these two methods
are limited by their inability to produce stable results.
Therefore, one of the design principles of our
system is to construct an algorithm that exhibits
both simplicity and stability when used for segmenting
the beef rib-eye image into lean and fat
regions based upon its characteristics. To achieve
this, we adopt the method of DTSM proposed by
Otsu (1979). The method of DTSM can globally
determine the threshold from the gray-level histogram
being generated from the beef rib-eye image.
Next, it is necessary to identify and to quantify
the features from beef rib-eye images so that beef
marbling can be graded in accordance with the
twelve BMS No. categories. In literatures presented