3.2.3. Extraction of features of beef marbling
Next, the features of beef marbling are extracted
from the binary image. Here, run length
processing is applied to count the number of successive
pixels according to lean and fat regions,
respectively, and store them in the order of
scanning. Tang (1998) introduced Run-Length
Run-Number Vector which represents the distribution
of the number of runs with a given run
length in the classification of texture images. Here,
we use the distribution of the run length ratio of
lean and fat runs over a transversal scan line on
the observation area in order for the feature
quantities of beef marbling to be invariant to the
size of input images.
Let the size of a run with bi;j 0 be R0p
, the size
of a run with bi;j 1 be R1p
, respectively, where p
and q represent the order of runs which are
counted in the scanned order. Fig. 5 illustrates the
run length processing. To get the distribution parameters
in particular, a beef rib-eye image is
separately scanned along x-axis and y-axis of the
vertical direction. The ratio of the size of lean run
at the Pth on a given scan line is denoted as
R0
P=
M 1 and R0
P =
N 1;
5
respectively. Similarly, the ratio of the size of fat
run at the Qth is denoted as
R1
Q=
M 1 and R1
Q=
N 1;
6
respectively. The histogram is being constructed,
where random variable is given by normalizing the
ratio over the size of run denoted by (5) and (6).
We call it run length histogram. In this paper, the
frequencies of run length ratio are decided to be
every 0:1 10ÿ3 in the domain of [0, 1]. The run
length histogram is being quantized based on the
number of divisions after dividing the domain of
[0,1] equidistantly, as shown in Fig. 6. We call this
normalized run length run number vectors
(NRLRNV). We adopt NRLRNV of lean and fat
regions as the features of beef marbling.
3.2.3. Extraction of features of beef marblingNext, the features of beef marbling are extractedfrom the binary image. Here, run lengthprocessing is applied to count the number of successivepixels according to lean and fat regions,respectively, and store them in the order ofscanning. Tang (1998) introduced Run-LengthRun-Number Vector which represents the distributionof the number of runs with a given runlength in the classification of texture images. Here,we use the distribution of the run length ratio oflean and fat runs over a transversal scan line onthe observation area in order for the featurequantities of beef marbling to be invariant to thesize of input images.Let the size of a run with bi;j 0 be R0p, the sizeof a run with bi;j 1 be R1p, respectively, where pand q represent the order of runs which arecounted in the scanned order. Fig. 5 illustrates therun length processing. To get the distribution parametersin particular, a beef rib-eye image isseparately scanned along x-axis and y-axis of thevertical direction. The ratio of the size of lean runat the Pth on a given scan line is denoted asR0P=
M 1 and R0P =
N 1;
5respectively. Similarly, the ratio of the size of fatrun at the Qth is denoted asR1Q=
M 1 and R1Q=
N 1;
6respectively. The histogram is being constructed,where random variable is given by normalizing theratio over the size of run denoted by (5) and (6).We call it run length histogram. In this paper, thefrequencies of run length ratio are decided to beevery 0:1 10ÿ3 in the domain of [0, 1]. The runlength histogram is being quantized based on thenumber of divisions after dividing the domain of[0,1] equidistantly, as shown in Fig. 6. We call thisnormalized run length run number vectors(NRLRNV). We adopt NRLRNV of lean and fatregions as the features of beef marbling.
การแปล กรุณารอสักครู่..
