6
BC images have more random, diverse, and heterogeneous
textures. As stated by Rashed et al. (2018)which are expen-
sive and time consuming. The research work outlined in
this paper examines the use of indirect strength estimation
methods-the point load test (PLT, BC has denser cleats than
BBC. It can be expected that, due to the denser presence of
cleats, there is more frequent change in the brightness in
the images of BC than those of BBC, potentially leading to
higher contrast and dissimilarity levels and lower correla-
tion, energy, and homogeneity levels. The results confirm
that the GLCM method can capture the differences in the
brightness profile between BC and BBC through textural
analyses. In addition, all the GLCM features for BC, except
the correlation along vertical direction, have a smaller
spread of values, indicating that the GLCM features for BC
are tightly clustered and the textures tend to be more con-
sistent than that of BBC.
Furthermore, the GLCM features were compared
between different directions. It can be found that, regard-
less of the coal lithotype, there is no substantial difference
in the GLCM features observed between the horizontal and
vertical directions. This suggests that the textural properties
of the images of BC and BBC exhibit a remarkable level of
symmetry or similarity along these two directions. Since
there are no significant differences in the GLCM features
calculated along the horizontal and vertical directions, the
GLCM features calculated along the horizontal direction
were used for all of the following analyses.
With the GLCM method, the distance refers to the
pixel distance between the reference pixel and the neighbor-
ing pixel for which co-occurrence is measured. Common
distance values include 1, 2 and 3 pixels. Different distances
measure different levels of texture information, and larger
distances capture longer-range texture patterns. Distance
values of 1, 2, and 3 were used to investigate the influence
of distance on the calculated GLCM features. Figure 4
shows the box plots of various GLCM features with dif-
ferent distance values, and the patch size is 50×50 pixels.
When altering the distance value, there are variations,
whether significant or subtle, observed in all the GLCM
features regardless of the coal lithotype. The trend in vari-
ations with distance is consistent for both BC and BBC
across all features. As the distance value increases, contrast
and dissimilarity increase. This is because larger distance
encompasses pixel pairs with greater intensity variations,
which are reflected in higher contrast and dissimilarity val-
ues. On the contrary, correlation, energy and homogeneity
decrease with the increase in distance value. As the distance
increases, the GLCM includes pixel pairs that are farther
apart, and the correlation tends to decrease because they are
less likely to exhibit strong linear dependencies at the same
time, there is a higher likelihood of encountering pixel pairs
with larger differences in gray levels, which can lead to the
decrease in homogeneity. Energy decreases or remains rela-
tively constant, indicating the texture becomes more het-
erogeneous or irregular with increasing distance.
In addition, each GLCM feature was compared
between BC and BBC. It can be found from the median
value and the spread of the data that, although there are still
differences in two groups of data, the distribution of these
groups of data become closer with the increase in distance
value. This makes them difficult to classify, which is not
preferable. Thus, a distance value of 1 was selected for the
following analyses.
Patch size is another factor affecting the calculated
GLCM features. The patch size determines the scale at
which texture features are extracted, and this can affect the
sensitivity to different texture patterns. Larger patch sizes
capture texture patterns at a coarser scale while smaller
patch sizes capture finer details. Smaller patches can be
more sensitive to noise, and larger patches can provide a
more stable texture analysis by averaging out noise. Larger
patches incorporate more contextual information, poten-
tially providing a better understanding of global texture
patterns, while smaller windows focus on local details.
Thus, it is a trade-off. Larger patches offer more context
Figure 4. Box plots of various GLCM features with different
distance values
BC images have more random, diverse, and heterogeneous
textures. As stated by Rashed et al. (2018)which are expen-
sive and time consuming. The research work outlined in
this paper examines the use of indirect strength estimation
methods-the point load test (PLT, BC has denser cleats than
BBC. It can be expected that, due to the denser presence of
cleats, there is more frequent change in the brightness in
the images of BC than those of BBC, potentially leading to
higher contrast and dissimilarity levels and lower correla-
tion, energy, and homogeneity levels. The results confirm
that the GLCM method can capture the differences in the
brightness profile between BC and BBC through textural
analyses. In addition, all the GLCM features for BC, except
the correlation along vertical direction, have a smaller
spread of values, indicating that the GLCM features for BC
are tightly clustered and the textures tend to be more con-
sistent than that of BBC.
Furthermore, the GLCM features were compared
between different directions. It can be found that, regard-
less of the coal lithotype, there is no substantial difference
in the GLCM features observed between the horizontal and
vertical directions. This suggests that the textural properties
of the images of BC and BBC exhibit a remarkable level of
symmetry or similarity along these two directions. Since
there are no significant differences in the GLCM features
calculated along the horizontal and vertical directions, the
GLCM features calculated along the horizontal direction
were used for all of the following analyses.
With the GLCM method, the distance refers to the
pixel distance between the reference pixel and the neighbor-
ing pixel for which co-occurrence is measured. Common
distance values include 1, 2 and 3 pixels. Different distances
measure different levels of texture information, and larger
distances capture longer-range texture patterns. Distance
values of 1, 2, and 3 were used to investigate the influence
of distance on the calculated GLCM features. Figure 4
shows the box plots of various GLCM features with dif-
ferent distance values, and the patch size is 50×50 pixels.
When altering the distance value, there are variations,
whether significant or subtle, observed in all the GLCM
features regardless of the coal lithotype. The trend in vari-
ations with distance is consistent for both BC and BBC
across all features. As the distance value increases, contrast
and dissimilarity increase. This is because larger distance
encompasses pixel pairs with greater intensity variations,
which are reflected in higher contrast and dissimilarity val-
ues. On the contrary, correlation, energy and homogeneity
decrease with the increase in distance value. As the distance
increases, the GLCM includes pixel pairs that are farther
apart, and the correlation tends to decrease because they are
less likely to exhibit strong linear dependencies at the same
time, there is a higher likelihood of encountering pixel pairs
with larger differences in gray levels, which can lead to the
decrease in homogeneity. Energy decreases or remains rela-
tively constant, indicating the texture becomes more het-
erogeneous or irregular with increasing distance.
In addition, each GLCM feature was compared
between BC and BBC. It can be found from the median
value and the spread of the data that, although there are still
differences in two groups of data, the distribution of these
groups of data become closer with the increase in distance
value. This makes them difficult to classify, which is not
preferable. Thus, a distance value of 1 was selected for the
following analyses.
Patch size is another factor affecting the calculated
GLCM features. The patch size determines the scale at
which texture features are extracted, and this can affect the
sensitivity to different texture patterns. Larger patch sizes
capture texture patterns at a coarser scale while smaller
patch sizes capture finer details. Smaller patches can be
more sensitive to noise, and larger patches can provide a
more stable texture analysis by averaging out noise. Larger
patches incorporate more contextual information, poten-
tially providing a better understanding of global texture
patterns, while smaller windows focus on local details.
Thus, it is a trade-off. Larger patches offer more context
Figure 4. Box plots of various GLCM features with different
distance values