Figure 5c: Image segmentation (Level 1, scale: 100) 
The classes created were built-up area, road (for 
identifying the larger roads that were not included in the 
built-up area), bare ground or sand, vegetation and water. 
Samples were selected for each class, and the image was 
classified using the Nearest Neighbour (NN) method. A 
large scale parameter was chosen for classifying the 
image in order to adequately represent the large built-up 
areas. The following features were used in the NN 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B7, 2012 
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia 
   
classification: mean values for red, green, blue and near 
infrared, brightness, maximum difference, compactness, 
length, length/width, HSI transformation, NDVI, GLCM 
mean (quick 8/11) (all dir.) and GLDV Ang. 2"! moment 
mean (quick 8/11) (all dir.). 
  
Figure 6: Object-based classification using the Nearest 
Neighbour method 
The classification results indicate that there is some 
confusion between certain classes, particularly where 
there are segments that contain more than one feature. 
This is a typical problem where large segments contain 
more than one feature class. On the contrary, smaller 
segments may represent individual features more easily, 
but the spectral differences within classes may result in 
the user having numerous sub-classes for features. 
  
  
  
  
  
  
  
  
  
  
  
Class name Producers Users KIA per 
accuracy accuracy class 
Built-up 0.57 0.80 0.48 
Road 1.00 1.00 1.00 
Vegetation 0.71 0.56 0.59 
Water 1.00 1.00 1.00 
Bare ground or 0.50 0.50 0.46 
sand 
Overall accuracy 0.79 
KIA 0.73 
  
  
  
Table 2: Accuracy assessment and kappa statistics for 
object-based classification (segments based on only 
spectral information) 
Using vector data and spectral information for 
segmentation 
In order to overcome the problem of having unsuitable 
segments such as those that spanned across roads or that 
contained mixed classes, vector data was included in 
order to segment the image based on cadastral 
information (information maintained and supplied by the 
Office of the Chief Surveyor General in South Africa). 
The initial segmentation was performed using the 
thematic layer to create segments at the cadastral layer 
level.