0 Benen
lass of the RDF
lh respect to the
ition accuracy of
ication accuracy
for training.
mean shift seg-
? confusion ma-
class (rows) and
indicate the true
(Pr).
vw
9 16
29 2
8553]
16 3
0.0
y. 7
76 5
4 68
esults. In Fig. 6,
s. For example,
) the reflectance
t sky regions are
ned label car in
ed simply by in-
., 2008), such as
below the build-
ded by building.
sification results
performance of
and RDF clas-
enefits from the
> feature sets we
ntation method,
Soille (1991), to
h the RDF clas-
e decision trees
n in Table 7.
h class remains
ge regions from
ind the low clas-
ither good fea-
Figure 5: Qualitative classification results of a RDF classifier
with the mean shift on the testing images from the eTRIMS
dataset. (Left: test image, middle: result, right: ground truth.)
Figure 6: Some more classification results of a RDF classifier
with the mean shift on the testing images from the eTRIMS
dataset. (Left: test image, middle: result, right: ground truth.)
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B3, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
Table 7: Pixelwise accuracy of the image classificationusing the
RDF classifier and the watershed segmentation on the eTRIMS
8-class dataset. The confusion matrix shows the classification
accuracy for each class (rows) and is row-normalized to sum to
100%. Row labels indicate the true class (Tr), and column labels
the predicted class (Pr).
Pr T b C d p T s v w
b 59 4 1 Br 5 9 ll 7
c 67 231 0 5 2 0 3 2
d 1950-12 5:0. 0:10. 002 417
p 57 3. 0-9: 3004040 1
r 14 1 0:58:23 1 3 1
s 17 50: 0 (16. 0:573. 2 1
v 13 4 | 2 1 13 61 4
w 23 1 1 1 0-6 .3 5]
5.4 Discussion
With respect to the three most important classes building, win-
dow, and vegetation, we are satisfied with our classification re-
sults. But our multi-class approach does not perform very well
for most of the other classes. Our classification scheme is faced
with a dramatic inequality between the sizes of the classes. Al-
most 6046 of the data is covered by only 2 classes, and the rest is
spread over the rest classes. And for the classes like car and door,
Gestalt features (Bileschi and Wolf, 2007) may play major role in
a good classification performance. We also believe symmetry and
repetition features are vital for classifying window class.
In this paper, features are extracted at local scale. Classification
results are achieved from bottom up on these local features by
classifiers. This factor leads to noisy boundaries in the example
images. To enforce consistency, a Markov or conditional ran-
dom field (Shotton et al., 2006) is often introduced for refinement,
which would likely improve the performance.
6 CONCLUSIONS
We evaluate the performance of seven feature sets with respect
to region-based classification of facade images. The feature sets
include basic features, color features, histogram features, Peucker
features, texture features, and SIFT features. We use randomized
decision forest (RDF) to perform the classification scheme. In our
experiments on the eTRIMS dataset (Koré and Fôrstner, 2009),
we have shown that RDF produces some reasonable classification
results.
The results show that these features and a local classifier are not
sufficient. In order to recover more precise boundaries, the work
presented in this paper has been fused into conditional random
field framework (Yang and Fórstner, 2011) by including neigh-
boring region information in the pairwise potential of the model,
which allows us to reduce misclassification that occurs near the
edges of objects. As future work, we are interested in evaluat-
ing more features, such as Gestalt features (Bileschi and Wolf,
2007) and other descriptor features (van de Sande et al., 2010),
for building facade images.
References
Barnard, K., Duygulu, P., Freitas, N. D., Forsyth, D., Blei, D. and
Jordan, M., 2003. Matching words and pictures. Journal of
Machine Learning Research 3, pp. 1107-1135.
