The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
spatial indexes to automatically maintain the topology between
parent and child parcels.
Image
Parcel number
Time consuming
Evaluation of segmentation precision
Our
method
eCognition
Our
method
eCognition
SPOT-5 multi-
spectral (687x569)
229
366
About
28s
About 8s
Both methods segment the urban,
rivers, ponds, mountains etc.
correctly, but our method better
keeps the boundary of the ponds.
IKONOS
panchromatic
(1142x787)
283
515
About
65s
About 15s
The segmentations of both methods
are totally similar. eCognition is
more regular in shapes, but with
fragmentized parcels distributing
around the boundary of many even-
tone, large-size parcels.
Google Earth
screen capture
(1208x796)
811
1778
About
58s
About 17s
Similar in segmentations. Both with
some errors. eCogntion is more
regular in shapes, but with the above
mentioned fragmentized parcels.
Table 1. Comparison of our method and the multi-resolution segmentation module of eCognition
Figure 1. Segmentation of a SPOT-5 image with scale 40
Figure 2. Segmentation of an Ikonos image with scale 80
1216