A Hybrid Dense Matching Method for Satellite Images

Abstract:

To solve the dense matching problem for stereoscopic satellite images, a hybrid matching scheme integrating multiple methods is proposed. This scheme utilizes two types of matching element including grid points and edge points. First the geometrically constrained cross-correlation (GC³) method is used to extract matching grids. While in less textured regions using GC³ cannot get sufficient matching grids, so the local affine transformation is used to establish the region correspondence, and more matching grids can be generated. Edges are extracted by Canny operator and approximated with a series of straight edge segments using a polygon approximation. Based on these approximated edges, edge correspondences between image pairs are established using GC³. This scheme fuses the region based and the feature based matching methods. It has been tested with real satellite images and the results demonstrate its accuracy and efficiency.

• Periodical: Advanced Materials Research (Volumes 726-731)
• Pages: 4547-4551
• DOI: https://doi.org/10.4028/www.scientific.net/AMR.726-731.4547
• Online since: August 2013
• Authors: Sheng Hua Teng*, Ning Yang
• Keywords: Affine Transformation, Edge Detection, Geometrically Constrained Cross-Correlation

References

[1] Suya You, Guangyou Xu: Journal of Image and Graphics, Vol. 2(1997), p.17 (In Chinese).

[2] Li Zhang: Automatic Digital Surface Model (DSM) Generation from Linear Array Images. Doctoral thesis of Swiss Federal Institute of Technology (2005).

[3] Canny J. F: IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 8(1986), p.679.

[4] Grimson W. E. L: Object Recognition by Computer (The MIT Press, USA 1990).