An Adaptive Active Contour Model for Building Extraction from Aerial Images

  • Mohammad Abdel Hameed Oudah -----> Dr. Ashraf Alattar

Building extraction from aerial images is one of the recent topics of remote sensing that
used in many applications such as urban planning, disaster management, military
planning, and Geographic Information Systems (GIS).
One of the most used approaches in building extraction is the Active Contour Model
(ACM) or snakes for its ability to extract contours of structured and unstructured shapes
of objects. However, using the traditional ACM snake model in building extraction and
other fields faces the problem of extracting contours of concavity regions, because snake
points cannot converge inside narrow concavity regions during its movement.
In this research we proposed to solve extraction contours of concavity regions problem
by adapting coefficients of ACM forces during snake iterations by adding a concavity
index to indicate that snake points stop in a concave region or not. Then adapt these
coefficients in term of concavity index value, to allow snake converge inside concavity
region.
Our adaptive model was tested on different sets of sub aerial images of buildings that
contain concavity regions, we show the results and evaluate these results using two
evaluation methods, the first evaluation method done in terms of accuracy, precision
and recall. And the second evaluation method evaluates the Error Distance Ratio (𝐸𝑅𝑑)
which is the average ratio of distance between each snake point and the true edge map
point (by pixels), the result values is compared with the GVF snake model, which is an
important improved ACM model that solved the concavity contour extraction problem.
In addition, we test and compare the execution time of our adaptive ACM model and
GVF model.