Interactive geospatial object extraction in high resolution remote sensing images using shape-based global minimization active contour model

• Combing edge, region and shape information for shape regular geospatial object extraction in remote sensing images.
• A new energy function based on active contour model is proposed.
• Shadow information is utilized to get more accurate extraction result.
• An iterative global minimization method is used to avoid the local minimum of the traditional level set method.
• Objects with complex structure e.g. aircraft and disturbing background are well extracted with our method.

In this work, we propose a novel algorithm to extract geospatial objects with regular shape in remote sensing images, using shape-based global minimization active contour model (SGACM). Specially, we define a new energy function combining both image appearance information and object shape prior, and minimize it with an iterative global minimization method. In the proposed energy, not only image edge and color information are utilized, but also a new shadow region term is introduced to obtain more accurate extraction result; moreover, a new shape energy term in which we use kernel principle component analysis (KPCA) to model shapes is defined in our method, which provides good constraint on the extraction process and makes results more robust with respect to disturbances. In the energy numerical minimization process, Split Bregman method is used to get a global solution which overcomes the drawback of running into local minimum for the traditional level set method. Experiment results demonstrate more robustness and accuracy of our proposed method compared with others without shape constraint.