Sander Oude Elberink
normalised DSM, i.e. the difference between DSM and DTM is often calculated as a first step (Haala, 1999) The
required DTM can be derived from the DSM by mathematical gray scale morphology like it is suggested by Weidner
and Förstner (1995). In their approach the DSM is processed by a morphological opening of the DSM Surface
eliminating all local maxima in height of a predefined size. In the normalised DSM buildings and trees rising from the
terrain approximately put on a plane, see figure 3. This is an important step to narrow the gray value cluster of building
and trees, and to make a proper discrimination between high and low objects.
3 TEXTURE IN LASER SCANNER DATA
Texture is an important characteristic for the segmentation of an image. In a laser scanner image texture is given by
local variation of height and height derivatives.
Here the aim of the use of texture measures is to discriminate between man-made and natural objects. In order t
classify buildings and trees in laser scanner data one can their difference in height texture. In general, buildings wj
show a regular, smooth pattern with small variations in height, while trees show an irregular pattern with high height
variations. When a small-area patch has wide variation of gray level primitives, the dominant property is texture
(Haralick and Shapiro, 1992). In this research texture features derived from co-occurrence matrices have been used t
detect trees in laser scanner data and afterwards make a proper discrimination between trees and buildings.
3.1 Co-occurrence matrix
Texture is characterized by its gray level primitive properties as well as the spatial relationships between them. The gray
level co-occurrence can be specified in a matrix of relative frequencies fij) with which two neighboring pixel;
separated by a distance d occur on the image, one with gray level i and the other with gray level j (Haralick and Shapiro,
1992), see figure 4.
—
t3
UU
M ha
Colin Che
uo c2 -] 0 ] 2 3 ue DON ; ur
29 0 $ f T101
Rai
2 02 20 2313 313
Aas gs 0 9. ie
3| 2 à 3 303 : Le frein
Tm
J
V JA AN. J
Y Y Y.
Image space Neighbourhood Co-occurrence matrix space
relationships
Fig. 4: Transformation of the gray value relationships within a 7 x 7 window into the co-occurrence matrix space with
horizontal and vertical directions (interpixel distance = 1) (modified after Zhang, 1999).
On the hand of the co-occurrence matrix several measures like correlation. entropy, contrast and angular second
moment can be calculated to discriminate between several textural objects.
3.2 Contrast texture measure in a laser scanner image
The main property of trees in an laser scanner image is the high local height variations, which are very well described
by the co-occurrence contrast texture measure as e.q. implemented in ENVI. Typical values of the parameters kappa
and lambda vary between 1 and 3.
680 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.