[R-sig-Geo] Feature Extraction - object based image analysis

[Tomislav Hengl]

Hi Barry,

SAGA has few simple but useful modules for extraction of features from grids. See:

> library(RSAGA)
> rsaga.get.modules("grid_discretisation")
$grid_discretisation
  code                       name interactive
1    0  Supervised Classification       FALSE
2    1 Cluster Analysis for Grids       FALSE
3    2          Grid Segmentation       FALSE
4    3      Grid Segmentation (b)       FALSE
5    4       Grid Skeletonization       FALSE
6   NA                       <NA>       FALSE
7   NA                       <NA>       FALSE


For multiple images, you can use either cluster analysis or supervised classification:

> rsaga.get.usage("grid_discretisation", 0)
SAGA CMD 2.0.3
library path:   C:/Progra~1/saga_vc/modules
library name:   grid_discretisation
module name :   Supervised Classification
Usage: 0 -GRIDS <str> -POLYGONS <str> [-FIELD <num>] -CLASSES <str> -RESULT <str> [-ML_PROB <str>]
[-METHOD <num>] [-NORMALISE] [-ML_THRESHOLD <str>]
  -GRIDS:<str>          Grids
        Grid list (input)
  -POLYGONS:<str>       Training Areas
        Shapes (input)
  -FIELD:<num>          Class Identifier
        Table field
  -CLASSES:<str>        Class Information
        Table (output)
  -RESULT:<str>         Classification
        Grid (output)
  -ML_PROB:<str>        Distance/Probability
        Grid (optional output)
  -METHOD:<num>         Method
        Choice
        Available Choices:
        [0] Minimum Distance
        [1] Maximum Likelihood
  -NORMALISE            Normalise
        Boolean
  -ML_THRESHOLD:<str>   Probability Threshold (Percent)
        Floating point
        Value Range: 0.000000 - 100.000000


You can also consider getting the statistics from grids for polygons:

> rsaga.get.modules("shapes_grid")
$shapes_grid
   code                             name interactive
1     0        Add Grid Values to Points       FALSE
2     1         Get Grid Data for Shapes       FALSE
3     2     Grid Statistics for Polygons       FALSE
4     3            Grid Values to Points       FALSE
5     4 Grid Values to Points (randomly)       FALSE
6     5          Contour Lines from Grid       FALSE
7     6         Vectorising Grid Classes       FALSE
8     7           Clip Grid with Polygon       FALSE
9     8               Gradient from Grid       FALSE
10   NA                             <NA>       FALSE
11   NA                             <NA>       FALSE


You will soon discover that there is not much literature that explains how to run processing and
what does each parameter in SAGA means  :( 
Unfortunately, Olaf does not have time to write up a user's manual for SAGA, but you can always zoom
into the original code. I am sure that all algorithms come from some literature. E.g.:

Kothe, R., Bock, M., 2006. Development and use in practice of SAGA modules for quality analysis of
geodata. In Böhner, J., McCloy, K.R., Strobl, J. [Eds.]: SAGA – Analysis and Modelling Applications.
Göttinger Geographische Abhandlungen, Vol.115.
http://surfnet.dl.sourceforge.net/sourceforge/saga-gis/gga115_08.pdf 


HTH

Tom Hengl
http://spatial-analyst.net  



-----Original Message-----
From: r-sig-geo-bounces at stat.math.ethz.ch [mailto:r-sig-geo-bounces at stat.math.ethz.ch] On Behalf Of
Rawlins, Barry G
Sent: Friday, February 27, 2009 3:06 PM
To: r-sig-geo at stat.math.ethz.ch
Subject: [R-sig-Geo] Feature Extraction - object based image analysis

I have a colleague who is not familiar with R and we were discussing feature extraction from
multiple images (air photos, satellite based etc.). He uses a commercial tool for feature
extraction. Can anyone point me to a package in R that does this? I have done a search but not found
such a function. It is not a procedure I am familiar with.

Many thanks, Barry

Dr Barry Rawlins
Sustainable Soils Team Leader
British Geological Survey
Keyworth
Nottingham
NG12 5GG
Direct tel: 0115 9363140
Mobile: 0788 4235473
homepage: http://barryrawlins.googlepages.com<http://barryrawlins.googlepages.com/>




-- 
This message (and any attachments) is for the recipient ...{{dropped:8}}