Geografic-Information-System/raster/r.texture/r.texture.html

<h2>DESCRIPTION</h2>

<em>r.texture</em> creates raster maps with textural features from a
user-specified raster map layer. The module calculates textural features 
based on spatial dependence matrices at 0, 45, 90, and 135 
degrees.

<p>
In order to take into account the scale of the texture to be measured,
<em>r.texture</em> allows the user to define the <em>size</em> of the moving
window and the <em>distance</em> at which to compare pixel grey values.  By
default the module averages the results over the 4 orientations, but the user
can also request output of the texture variables in 4 different orientations
(flag <em>-s</em>). Please note that angles are defined in degrees of east and
they increase counterclockwise, so 0 is East - West, 45 is North-East -
South-West, 90 is North - South, 135 is North-West - South-East.

<p>
The user can either chose one or several texture measures (see below for their
description) using the <em>method</em> parameter, or can request the creating
of maps for all available methods with the <em>-a</em>.

<p>
<em>r.texture</em> assumes grey levels ranging from 0 to 255 as input.  The
input is automatically rescaled to 0 to 255 if the input map range is outside of
this range.  In order to reduce noise in the input data (thus generally
reinforcing the textural features), and to speed up processing, it is
recommended that the user recode the data using equal-probability quantization.
Quantization rules for <em>r.recode</em> can be generated with <em>r.quantile
-r</em> using e.g 16 or 32 quantiles (see example below).


<h2>NOTES</h2>

<p>
Texture is a feature of specific land cover classes in satellite imagery.
It is particularly useful in situations where spectral differences between
classes are small, but classes are distinguishable by their organisation on the 
ground, often opposing natural to human-made spaces: cultivated fields vs meadows
or golf courses, palm tree plantations vs natural rain forest, but texture can
also be a natural phenomen: dune fields, different canopies due to different
tree species. The usefulness and use of texture is highly dependent on the 
resolution of satellite imagery and on the scale of the human intervention or 
the phenomenon that created the texture (also see the discussion of scale 
dependency below). The user should observe the phenomenon visually in order to
determine an adequat setting of the <em>size</em> parameter.

<p>
The output of <em>r.texture</em> can constitute very useful additional variables 
as input for image classification or image segmentation (object recognition). 
It can be used in supervised classification algorithms such as 
<a href="i.maxlik.html">i.maxlik</a> or <a href="i.smap.html">i.smap</a>,
or for the identification of objects in <a href="i.segment.html">i.segment</a>,
and/or for the characterization of these objects and thus, for example, as one 
of the raster inputs of the 
<a href="https://grass.osgeo.org/grass7/manuals/addons/i.segment.stats.html">
	i.segment.stats</a> addon.

<p>
In general, several variables constitute texture: differences in grey level values,
coarseness as scale of grey level differences, presence or lack of directionality
and regular patterns. A texture can be characterized by tone (grey level intensity
properties) and structure (spatial relationships). Since textures are highly scale
dependent, hierarchical textures may occur.

<p>
<em>r.texture</em> uses the common texture model based on the so-called grey 
level co-occurrence matrix as described by Haralick et al (1973). This matrix 
is a two-dimensional histogram of grey levels for a pair of pixels which are 
separated by a fixed spatial relationship. The matrix approximates the joint 
probability distribution of a pair of pixels. Several texture measures are 
directly computed from the grey level co-occurrence matrix. 

<p>
The following part offers brief explanations of the Haralick et al texture 
measures (after Jensen 1996).

<h3>First-order statistics in the spatial domain</h3>
<ul>
<li> Sum Average (SA)</li>

<li> Entropy (ENT):
 This measure analyses the randomness. It is high when the values of the
 moving window have similar values. It is low when the values are close
 to either 0 or 1 (i.e. when the pixels in the local window are uniform).</li>

<li> Difference Entropy (DE)</li>

<li> Sum Entropy (SE)</li>

<li> Variance (VAR):
  A measure of gray tone variance within the moving window (second-order
moment about the mean)</li>

<li> Difference Variance (DV)</li>

<li> Sum Variance (SV)</li>
</ul>

Note that measures "mean", "kurtosis", "range", "skewness", and "standard
deviation" are available in <em>r.neighbors</em>.

<h3>Second-order statistics in the spatial domain</h3>

The second-order statistics texture model is based on the so-called grey
level co-occurrence matrices (GLCM; after Haralick 1979).

<ul>
<li> Angular Second Moment (ASM, also called Uniformity):
 This is a measure of local homogeneity and the opposite of Entropy.
 High values of ASM occur when the pixels in the moving window are
 very similar.
 <br>
 Note: The square root of the ASM is sometimes used as a texture measure,
 and is called Energy.</li>

<li> Inverse Difference Moment (IDM, also called Homogeneity):
 This measure relates inversely to the contrast measure. It is a direct measure of the
 local homogeneity of a digital image. Low values are associated with low homogeneity
 and vice versa.</li>

<li> Contrast (CON):
 This measure analyses the image contrast (locally gray-level variations) as
 the linear dependency of grey levels of neighboring pixels (similarity). Typically high,
 when the scale of local texture is larger than the <em>distance</em>.</li>

<li> Correlation (COR):
 This measure  analyses the linear dependency of grey levels of neighboring
 pixels. Typically high, when the scale of local texture is larger than the
 <em>distance</em>.</li>

<li> Information Measures of Correlation (MOC)</li>

<li> Maximal Correlation Coefficient (MCC)</li>
</ul>

<p>
The computational region should be set to the input map with 
<b>g.region raster=&lt;input map&gt;</b>, or aligned to the input map 
with <b>g.region align=&lt;input map&gt;</b> if only a subregion 
should be analyzed.

<p>
Note that the output of <em>r.texture</em> will always be smaller than 
the current region as only cells for which there are no null cells and 
for which all cells of the moving window are within the current region 
will contain a value. The output will thus appear cropped at the margins.

<p>
Importantly, the input raster map cannot have more than 255 categories.

<h2>EXAMPLE</h2>

Calculation of Angular Second Moment of B/W orthophoto (North Carolina data set):

<div class="code"><pre>
g.region raster=ortho_2001_t792_1m -p
# set grey level color table 0% black 100% white
r.colors ortho_2001_t792_1m color=grey
# extract grey levels
r.mapcalc "ortho_2001_t792_1m.greylevel = ortho_2001_t792_1m"
# texture analysis
r.texture ortho_2001_t792_1m.greylevel prefix=ortho_texture method=asm -s 
# display
g.region n=221461 s=221094 w=638279 e=638694
d.shade color=ortho_texture_ASM_0 shade=ortho_2001_t792_1m
</pre></div>

This calculates four maps (requested texture at four orientations):
ortho_texture_ASM_0, ortho_texture_ASM_45, ortho_texture_ASM_90, ortho_texture_ASM_135.

Reducing the number of gray levels (equal-probability quantizing):

<div class="code"><pre>
g.region -p raster=ortho_2001_t792_1m

# enter as one line or with \
r.quantile input=ortho_2001_t792_1m quantiles=16 -r | r.recode \
           input=ortho_2001_t792_1m output=ortho_2001_t792_1m_q16 rules=-
</pre></div>

The recoded raster map can then be used as input for <em>r.texture</em> as before.

<p>
Second example: analysis of IDM (homogeneity) on a simple raster with 
North-South line pattern.

<div class="code"><pre>
# import raster
r.in.ascii in=- output=lines &lt;&lt; EOF
north: 9
south: 0
east: 9
west: 0
rows: 9
cols: 9
0 0 0 1 0 0 0 1 0
0 0 0 1 0 0 0 1 0
0 0 0 1 0 0 0 1 0
0 0 0 1 0 0 0 1 0
0 0 0 1 0 0 0 1 0
0 0 0 1 0 0 0 1 0
0 0 0 1 0 0 0 1 0
0 0 0 1 0 0 0 1 0
0 0 0 1 0 0 0 1 0
EOF

# adjust region to raster
g.region raster=lines

# calculate IDM (homogeneity) in all directions
r.texture -s lines method=idm output=text_lines
</pre></div>

<p>
The following image shows the original map, the result in East-West direction
and the result in North-South direction, showing how texture can depend on
direction, with texture perfectly homogeneous (value=1) in the North-South 
direction, but quite heterogeneous in East-West direction, except for those 
areas where there are three columns of equal values (as size=3). 
The overlaid grid highlights that the texture measures output maps
are cropped at the margins.

<center>
	<img src="r_texture_directions_example.png" border=1><br>
	<i>IDM textures according to direction</i>
</center>


<h2>KNOWN ISSUES</h2>
The program can run incredibly slow for large raster maps and large 
moving windows (<em>size</em> option).

<h2>REFERENCES</h2>

The algorithm was implemented after Haralick et al., 1973 and 1979.

<p>
The original code was taken by permission from <em>pgmtexture</em>, part of
PBMPLUS (Copyright 1991, Jef Poskanser and Texas Agricultural Experiment
Station, employer for hire of James Darrell McCauley). Manual page 
of <a href="http://netpbm.sourceforge.net/doc/pgmtexture.html">pgmtexture</a>.
Over the years, the source code of <em>r.texture</em> was further improved.

<ul> 
<li>Haralick, R.M., K. Shanmugam, and I. Dinstein (1973). Textural features for
    image classification. <em>IEEE Transactions on Systems, Man, and
    Cybernetics</em>, SMC-3(6):610-621.</li>
<li>Bouman, C. A., Shapiro, M. (1994). A Multiscale Random Field Model for
 Bayesian Image Segmentation, IEEE Trans. on Image Processing, vol. 3, no. 2.</li>
<li>Jensen, J.R. (1996). Introductory digital image processing. Prentice Hall.
  ISBN 0-13-205840-5 </li>
<li>Haralick, R. (May 1979). <i>Statistical and structural approaches to texture</i>,
   Proceedings of the IEEE, vol. 67, No.5, pp. 786-804</li>
<li>Hall-Beyer, M. (2007). <a href="http://www.fp.ucalgary.ca/mhallbey/tutorial.htm">The GLCM Tutorial Home Page</a>
  (Grey-Level Co-occurrence Matrix texture measurements). University of Calgary, Canada
</ul>

<h2>SEE ALSO</h2>

<em>
<a href="i.maxlik.html">i.maxlik</a>,
<a href="i.gensig.html">i.gensig</a>,
<a href="i.smap.html">i.smap</a>,
<a href="i.gensigset.html">i.gensigset</a>,
<a href="https://grass.osgeo.org/grass7/manuals/addons/i.segment.stats.html">
	        i.segment.stats</a>,
<a href="i.pca.html">i.pca</a>,
<a href="r.neighbors.html">r.neighbors</a>,
<a href="r.rescale.html">r.rescale</a>
</em>

<h2>AUTHORS</h2>
<a href="mailto:antoniol@ieee.org">G. Antoniol</a> - RCOST (Research Centre on Software Technology - Viale Traiano - 82100 Benevento)<br>
C. Basco -  RCOST (Research Centre on Software Technology - Viale Traiano - 82100 Benevento)<br>
M. Ceccarelli - Facolta di Scienze, Universita del Sannio, Benevento<br>
Markus Metz (correction and optimization of the initial version)<br>
Moritz Lennert (documentation)

<p><i>Last changed: $Date$</i>