Information Technology Reference
In-Depth Information
coarseness, and directionality. It is extracted from the frequency layout as
shown in Figure 11.1 at different scales decomposed using Gabor wave-
let. The dominant direction is based on directional histograms computed
from 4 × 6 filtered images that are used to construct four directional his-
tograms. Each histogram essentially captures the first-order statistics of
the distribution of edge pixels over
K
= 6 directions. The scale is computed
based on contrast of projections along two dominant orientations. The
regularity is computed by assigning credits to those rejections that pass a
consistency check as the periodicity of the basic texture elements, called
textons.
The texture browsing descriptor (TBD) is not a typical similarity matching
descriptor. It is used for browsing images. One can select any of the compo-
nents and browse along that dimension.
11.2.2.3 Edge Histogram Descriptor (EHD)
The EHD specifies the spatial distribution of five types of edges. It extracted
small square image blocks in each of the 16 nonoverlapping subimages. The
directional edge information is obtained from mean values for four sub-
blocks of image blocks. If the maximum of the five directional edge strengths
is greater than a thresholding value, the block is decided to have the corre-
sponding edge type.
The EHD captures spatial distribution of edges in images that are nonho-
mogeneous texture.
The similarity matching for EHD can be measured by the following
function
79
∑
|
D
=
Local_Edge
[ ]
i
−
ocal_Edge
[ ]|
i
1
2
i
=
0
4
∑
+×
5
|
Global_Edge
[]
i
−
lobal_Edge
[]|
i
(11.15)
1
2
i
=
0
64
∑
+
|
Sem
i_Global_Edge
[]
i
−
Semi_Global_Edge
[]|
i
,
1
2
i
=
0
where the Global_Edge histogram and Semi_Global edge histogram are gen-
erated from the 80 local edge histograms in the descriptor. The Global_Edge
histogram is obtained by accumulating and normalizing the bin values of
the corresponding edge type of the local edge histogram while the histo-
gram is obtained by the grouping subsets of the local edge histogram as
shown in Figure 11.2.
