5
GLCM features with the same camera settings
The two coal images shown in Figure 2 were used for the
analysis in this section. They were shot with the same cam-
era settings with only one difference, which is the focal
length, the distance between camera and the coal ribs.
Capturing an image from a closer distance will cause the
scene or object to occupy more pixels, resulting in a higher
resolution with finer details. GLCM is a texture analysis
method that depends on the spatial relationships between
pixel values in the image. When the coal images of differ-
ent resolutions are used, the spatial relationships between
pixel values change, and different GLCM features can be
obtained. Therefore, it is necessary to resize the images to
the same resolution for comparison. In a previous study (Yu
et al., 1997)methods for its characterisation and analysis are
poorly developed. Banding texture was obtained manually
from the coal face and core at a minimum resolution of
1 mm. Window filtering was used to determine the opti-
mum resolution (30–50 mm, coal rib images were used to
classify coal lithotypes and it was found that a minimum
resolution of 1 pixel/mm (25 pixels/in) is necessary for
the successful classification. Thus, all the coal images were
resized based on this resolution for the analyses in this sec-
tion. The detailed investigation of the influence of resolu-
tion on GLCM statics will be presented in next section.
Rather than using the whole image as one input for
GLCM calculation, small patches or sliding windows were
randomly extracted from the original image and were used
for GLCM calculation. Three patch sizes are used, namely
25×25 pixels, 50×50 pixels, 100×100 pixels, correspond-
ing to the covered areas of 25×25 mm, 50×50 mm, and
100×100 mm, respectively. For each patch size, 100 patches
were randomly extracted from each image in Figure 2. In
such a way, a database of GLCM features with different
coal lithotypes can be generated for statistical analyses.
Besides patch size, there are other parameters, namely angle
(or direction) and distance (or offset), potentially affect-
ing the calculated GLCM features. The influence of these
parameters on GLCM features of coal images of different
lithotypes are studied in this section.
With GLCM method, the angle specifies the direc-
tion in which the co-occurrence is calculated. Common
angles are 0 degree (horizontal), 45 degrees (diagonal), 90
degrees (vertical) and 135 degrees (the opposite diagonal).
Different angles capture texture patterns along different
directions. The box plot of various GLCM features along
horizontal (0 degree) and vertical (90 degree) directions is
shown in Figure 3, and all the other parameters were kept
the same. The patch size is 50×50 pixels, and a distance
value of 1 is used. First of all, the box plots were compared
between BC and BBC. It can be found that, for any GLCM
feature, there are observable differences in the median val-
ues. Compared to BBC, the textures of BC have higher
contrast and dissimilarity levels on average, indicating that
the image texture of BC has large local variation in pixel
intensities, resulting in sharp transitions. At the same time,
the textures of BC show generally lower correlation, energy,
and homogeneity levels than those of BBC, indicating that
Figure 2. Images of coal with different lithotypes extracted
from rib photos
Figure 3. Box plots of various GLCM features along two
different directions (horizontal and vertical)
GLCM features with the same camera settings
The two coal images shown in Figure 2 were used for the
analysis in this section. They were shot with the same cam-
era settings with only one difference, which is the focal
length, the distance between camera and the coal ribs.
Capturing an image from a closer distance will cause the
scene or object to occupy more pixels, resulting in a higher
resolution with finer details. GLCM is a texture analysis
method that depends on the spatial relationships between
pixel values in the image. When the coal images of differ-
ent resolutions are used, the spatial relationships between
pixel values change, and different GLCM features can be
obtained. Therefore, it is necessary to resize the images to
the same resolution for comparison. In a previous study (Yu
et al., 1997)methods for its characterisation and analysis are
poorly developed. Banding texture was obtained manually
from the coal face and core at a minimum resolution of
1 mm. Window filtering was used to determine the opti-
mum resolution (30–50 mm, coal rib images were used to
classify coal lithotypes and it was found that a minimum
resolution of 1 pixel/mm (25 pixels/in) is necessary for
the successful classification. Thus, all the coal images were
resized based on this resolution for the analyses in this sec-
tion. The detailed investigation of the influence of resolu-
tion on GLCM statics will be presented in next section.
Rather than using the whole image as one input for
GLCM calculation, small patches or sliding windows were
randomly extracted from the original image and were used
for GLCM calculation. Three patch sizes are used, namely
25×25 pixels, 50×50 pixels, 100×100 pixels, correspond-
ing to the covered areas of 25×25 mm, 50×50 mm, and
100×100 mm, respectively. For each patch size, 100 patches
were randomly extracted from each image in Figure 2. In
such a way, a database of GLCM features with different
coal lithotypes can be generated for statistical analyses.
Besides patch size, there are other parameters, namely angle
(or direction) and distance (or offset), potentially affect-
ing the calculated GLCM features. The influence of these
parameters on GLCM features of coal images of different
lithotypes are studied in this section.
With GLCM method, the angle specifies the direc-
tion in which the co-occurrence is calculated. Common
angles are 0 degree (horizontal), 45 degrees (diagonal), 90
degrees (vertical) and 135 degrees (the opposite diagonal).
Different angles capture texture patterns along different
directions. The box plot of various GLCM features along
horizontal (0 degree) and vertical (90 degree) directions is
shown in Figure 3, and all the other parameters were kept
the same. The patch size is 50×50 pixels, and a distance
value of 1 is used. First of all, the box plots were compared
between BC and BBC. It can be found that, for any GLCM
feature, there are observable differences in the median val-
ues. Compared to BBC, the textures of BC have higher
contrast and dissimilarity levels on average, indicating that
the image texture of BC has large local variation in pixel
intensities, resulting in sharp transitions. At the same time,
the textures of BC show generally lower correlation, energy,
and homogeneity levels than those of BBC, indicating that
Figure 2. Images of coal with different lithotypes extracted
from rib photos
Figure 3. Box plots of various GLCM features along two
different directions (horizontal and vertical)