segmSpannWilson
[ XITE Reference Manual | XITE home ]
Contents
Name
segmSpannWilson - segmentation by the Spann and Wilson method
#include <xite/segmentation.h>
BiffStatus segmSpannWilson( IBAND inband,
IBAND outband, int n, int m, int method );
Performs segmentation of the input band into the
output band. The algorithm has three steps:
-
Quad-tree smoothing
-
A quad-tree is created, and pixels are averaged 4 by 4
up in the quad-tree. n quad-tree levels are created,
including the bottom level, which is inband.
-
Local centroid clustering
-
The topmost quad-tree level is clustered by moving histogram
mass to the centroid (mean or median) of a local window moved
over the histogram. Applied iteratively, until stable. See
thresMlCentroid.
-
Boundary estimation
-
The classification result is copied down the quad-tree, but at
every level, all pixels close to a differently classified pixel
are reclassified to determine the border more precisely.
-
inband
- Input band to be segmented.
-
outband
- The result of the segmentation.
-
n
- Levels in the quad-tree (must be at least two).
-
m
- Size of histogram window, like 10-20.
-
method
- Local centriod clustering: 0 -> mean, 1 -> median.
Note that if you specify histogram window size m=0, the
algorithm will try to find an m-value where the clustering
result is relatively stable over a range of m-values.
inband and outband must have pixel type unsigned byte.
outband should have the same size as inband. Only the
largest common rectangle of inband and outband is handled.
segmSpannWilson(1), thresMlCentroid(3), segmRandom(3)
-
M. Spann and R. Wilson,
- "A Quad-Tree Approach to Image Segmentation which Combines
Statistical and Spatial Information"
Patteren Recognition, Vol 18, Nos 3/4, pp 257-269, 1985
0 => ok
2 => error: bad inband pixel type
3 => error: bad outband pixel type
4 => error: bad n value
5 => error: bad m value
6 => error: bad method value
Tor Lønnestad, BLAB, Ifi, UiO
$Id: segmSpannWilson.c,v 1.25 1997/01/16 16:08:53 svein Exp $