Line-network extraction

 


The objective of this problem consists in extracting line-networks from images. Different types of approaches are used in the litterature, mainly pixel-based, object-based and graph-based. Here, we aim to extract the structure of the networks which means we do not search for a binary classification roads/background at pixel level, but for an object-oriented representation.


Evaluation of the results: In its current version, results are evaluated through the precision of line-networks, the complexity of the representation, and time. You can use this Matlab code to evaluate your results from the following datasets.



 

Road
Acknowledgment: IGN


Image (894*650 pixels)
Ground truth

Results


 

River
Acknowledgment: BRGM


Image (364*320 pixels)
Ground truth

Results

 

 

Tiles


Image (675*523 pixels)
Ground truth

Results


 

Leaf


Image (651*364 pixels)
Ground truth

Results

 


 


back