1r.texture(1) GRASS GIS User's Manual r.texture(1)
2
6 r.texture - Generate images with textural features from a raster map.
7
9 raster, algebra, statistics, texture
10
12 r.texture
13 r.texture --help
14 r.texture [-san] input=name output=basename [size=value] [dis‐
15 tance=value] [method=string[,string,...]] [--overwrite] [--help]
16 [--verbose] [--quiet] [--ui]
17 Flags:
19 -s
20 Separate output for each angle (0, 45, 90, 135)
21 Angles are counterclockwise from east: 0 is East to West, 45 is
22 North-East to South-West
23 -a
25 Calculate all textural measurements
26 -n
28 Allow NULL cells in a moving window
29 This will also avoid cropping along edges of the current region
30 --overwrite
32 Allow output files to overwrite existing files
33 --help
35 Print usage summary
36 --verbose
38 Verbose module output
39 --quiet
41 Quiet module output
42 --ui
44 Force launching GUI dialog
45 Parameters:
47 input=name [required]
48 Name of input raster map
49 output=basename [required]
51 Name for output basename raster map(s)
52 size=value
54 The size of moving window (odd and >= 3)
55 Default: 3
56 distance=value
58 The distance between two samples (>= 1)
59 The distance must be smaller than the size of the moving window
60 Default: 1
61 method=string[,string,...]
63 Textural measurement method
64 Options: asm, contrast, corr, var, idm, sa, sv, se, entr, dv, de,
65 moc1, moc2
66
68 r.texture creates raster maps with textural features from a user-speci‐
69 fied raster map layer. The module calculates textural features based on
70 spatial dependence matrices at 0, 45, 90, and 135 degrees.
71 In order to take into account the scale of the texture to be measured,
73 r.texture allows the user to define the size of the moving window and
74 the distance at which to compare pixel grey values. By default the
75 module averages the results over the 4 orientations, but the user can
76 also request output of the texture variables in 4 different orienta‐
77 tions (flag -s). Please note that angles are defined in degrees of east
78 and they increase counterclockwise, so 0 is East - West, 45 is
79 North-East - South-West, 90 is North - South, 135 is North-West -
80 South-East.
81 The user can either chose one or several texture measures (see below
83 for their description) using the method parameter, or can request the
84 creating of maps for all available methods with the -a.
85 r.texture assumes grey levels ranging from 0 to 255 as input. The in‐
87 put is automatically rescaled to 0 to 255 if the input map range is
88 outside of this range. In order to reduce noise in the input data
89 (thus generally reinforcing the textural features), and to speed up
90 processing, it is recommended that the user recode the data using
91 equal-probability quantization. Quantization rules for r.recode can be
92 generated with r.quantile -r using e.g 16 or 32 quantiles (see example
93 below).
94
96 Texture is a feature of specific land cover classes in satellite im‐
97 agery. It is particularly useful in situations where spectral differ‐
98 ences between classes are small, but classes are distinguishable by
99 their organisation on the ground, often opposing natural to human-made
100 spaces: cultivated fields vs meadows or golf courses, palm tree planta‐
101 tions vs natural rain forest, but texture can also be a natural phe‐
102 nomen: dune fields, different canopies due to different tree species.
103 The usefulness and use of texture is highly dependent on the resolution
104 of satellite imagery and on the scale of the human intervention or the
105 phenomenon that created the texture (also see the discussion of scale
106 dependency below). The user should observe the phenomenon visually in
107 order to determine an adequat setting of the size parameter.
108 The output of r.texture can constitute very useful additional variables
110 as input for image classification or image segmentation (object recog‐
111 nition). It can be used in supervised classification algorithms such
112 as i.maxlik or i.smap, or for the identification of objects in i.seg‐
113 ment, and/or for the characterization of these objects and thus, for
114 example, as one of the raster inputs of the i.segment.stats addon.
115 In general, several variables constitute texture: differences in grey
117 level values, coarseness as scale of grey level differences, presence
118 or lack of directionality and regular patterns. A texture can be char‐
119 acterized by tone (grey level intensity properties) and structure (spa‐
120 tial relationships). Since textures are highly scale dependent, hierar‐
121 chical textures may occur.
122 r.texture uses the common texture model based on the so-called grey
124 level co-occurrence matrix as described by Haralick et al (1973). This
125 matrix is a two-dimensional histogram of grey levels for a pair of pix‐
126 els which are separated by a fixed spatial relationship. The matrix ap‐
127 proximates the joint probability distribution of a pair of pixels. Sev‐
128 eral texture measures are directly computed from the grey level co-oc‐
129 currence matrix.
130 The following part offers brief explanations of the Haralick et al tex‐
132 ture measures (after Jensen 1996).
133 First-order statistics in the spatial domain
135 • Sum Average (SA)
136 • Entropy (ENT): This measure analyses the randomness. It is high
138 when the values of the moving window have similar values. It is
139 low when the values are close to either 0 or 1 (i.e. when the
140 pixels in the local window are uniform).
141 • Difference Entropy (DE)
143 • Sum Entropy (SE)
145 • Variance (VAR): A measure of gray tone variance within the mov‐
147 ing window (second-order moment about the mean)
148 • Difference Variance (DV)
150 • Sum Variance (SV)
152 Note that measures "mean", "kurtosis", "range", "skewness", and "stan‐
153 dard deviation" are available in r.neighbors.
154 Second-order statistics in the spatial domain
156 The second-order statistics texture model is based on the so-called
157 grey level co-occurrence matrices (GLCM; after Haralick 1979).
158 • Angular Second Moment (ASM, also called Uniformity): This is a
160 measure of local homogeneity and the opposite of Entropy. High
161 values of ASM occur when the pixels in the moving window are
162 very similar.
163 Note: The square root of the ASM is sometimes used as a texture
164 measure, and is called Energy.
165 • Inverse Difference Moment (IDM, also called Homogeneity): This
167 measure relates inversely to the contrast measure. It is a di‐
168 rect measure of the local homogeneity of a digital image. Low
169 values are associated with low homogeneity and vice versa.
170 • Contrast (CON): This measure analyses the image contrast (lo‐
172 cally gray-level variations) as the linear dependency of grey
173 levels of neighboring pixels (similarity). Typically high, when
174 the scale of local texture is larger than the distance.
175 • Correlation (COR): This measure analyses the linear dependency
177 of grey levels of neighboring pixels. Typically high, when the
178 scale of local texture is larger than the distance.
179 • Information Measures of Correlation (MOC)
181 • Maximal Correlation Coefficient (MCC)
183 The computational region should be set to the input map with g.region
185 raster=<input map>, or aligned to the input map with g.region
186 align=<input map> if only a subregion should be analyzed.
187 Note that the output of r.texture will always be smaller than the cur‐
189 rent region as only cells for which there are no null cells and for
190 which all cells of the moving window are within the current region will
191 contain a value. The output will thus appear cropped at the margins.
192 Importantly, the input raster map cannot have more than 255 categories.
194
196 Calculation of Angular Second Moment of B/W orthophoto (North Carolina
197 data set):
198 g.region raster=ortho_2001_t792_1m -p
199 # set grey level color table 0% black 100% white
200 r.colors ortho_2001_t792_1m color=grey
201 # extract grey levels
202 r.mapcalc "ortho_2001_t792_1m.greylevel = ortho_2001_t792_1m"
203 # texture analysis
204 r.texture ortho_2001_t792_1m.greylevel prefix=ortho_texture method=asm -s
205 # display
206 g.region n=221461 s=221094 w=638279 e=638694
207 d.shade color=ortho_texture_ASM_0 shade=ortho_2001_t792_1m
208 This calculates four maps (requested texture at four orientations): or‐
209 tho_texture_ASM_0, ortho_texture_ASM_45, ortho_texture_ASM_90, or‐
210 tho_texture_ASM_135. Reducing the number of gray levels (equal-proba‐
211 bility quantizing):
212 g.region -p raster=ortho_2001_t792_1m
213 # enter as one line or with \
214 r.quantile input=ortho_2001_t792_1m quantiles=16 -r | r.recode \
215 input=ortho_2001_t792_1m output=ortho_2001_t792_1m_q16 rules=-
216 The recoded raster map can then be used as input for r.texture as be‐
217 fore.
218 Second example: analysis of IDM (homogeneity) on a simple raster with
220 North-South line pattern.
221 # import raster
222 r.in.ascii in=- output=lines << EOF
223 north: 9
224 south: 0
225 east: 9
226 west: 0
227 rows: 9
228 cols: 9
229 0 0 0 1 0 0 0 1 0
230 0 0 0 1 0 0 0 1 0
231 0 0 0 1 0 0 0 1 0
232 0 0 0 1 0 0 0 1 0
233 0 0 0 1 0 0 0 1 0
234 0 0 0 1 0 0 0 1 0
235 0 0 0 1 0 0 0 1 0
236 0 0 0 1 0 0 0 1 0
237 0 0 0 1 0 0 0 1 0
238 EOF
239 # adjust region to raster
240 g.region raster=lines
241 # calculate IDM (homogeneity) in all directions
242 r.texture -s lines method=idm output=text_lines
243 The following image shows the original map, the result in East-West di‐
245 rection and the result in North-South direction, showing how texture
246 can depend on direction, with texture perfectly homogeneous (value=1)
247 in the North-South direction, but quite heterogeneous in East-West di‐
248 rection, except for those areas where there are three columns of equal
249 values (as size=3). The overlaid grid highlights that the texture mea‐
250 sures output maps are cropped at the margins.
251 IDM textures according to direction
252
254 The program can run incredibly slow for large raster maps and large
255 moving windows (size option).
256
258 The algorithm was implemented after Haralick et al., 1973 and 1979.
259 The original code was taken by permission from pgmtexture, part of PBM‐
261 PLUS (Copyright 1991, Jef Poskanser and Texas Agricultural Experiment
262 Station, employer for hire of James Darrell McCauley). Manual page of
263 pgmtexture. Over the years, the source code of r.texture was further
264 improved.
265 • Haralick, R.M., K. Shanmugam, and I. Dinstein (1973). Textural
267 features for image classification. IEEE Transactions on Sys‐
268 tems, Man, and Cybernetics, SMC-3(6):610-621.
269 • Bouman, C. A., Shapiro, M. (1994). A Multiscale Random Field
271 Model for Bayesian Image Segmentation, IEEE Trans. on Image
272 Processing, vol. 3, no. 2.
273 • Jensen, J.R. (1996). Introductory digital image processing.
275 Prentice Hall. ISBN 0-13-205840-5
276 • Haralick, R. (May 1979). Statistical and structural approaches
278 to texture, Proceedings of the IEEE, vol. 67, No.5, pp. 786-804
279 • Hall-Beyer, M. (2007). The GLCM Tutorial Home Page (Grey-Level
281 Co-occurrence Matrix texture measurements). University of Cal‐
282 gary, Canada
283
285 i.maxlik, i.gensig, i.smap, i.gensigset, i.segment.stats, i.pca,
286 r.neighbors, r.rescale
287
289 G. Antoniol - RCOST (Research Centre on Software Technology - Viale
290 Traiano - 82100 Benevento)
291 C. Basco - RCOST (Research Centre on Software Technology - Viale Tra‐
292 iano - 82100 Benevento)
293 M. Ceccarelli - Facolta di Scienze, Universita del Sannio, Benevento
294 Markus Metz (correction and optimization of the initial version)
295 Moritz Lennert (documentation)
296
298 Available at: r.texture source code (history)
299 Accessed: Mon Jun 20 16:46:33 2022
301 Main index | Raster index | Topics index | Keywords index | Graphical
303 index | Full index
304 © 2003-2022 GRASS Development Team, GRASS GIS 8.2.0 Reference Manual
306GRASS 8.2.0 r.texture(1)