A Quick Image Registration Algorithm Based on Delaunay Triangulation

The traditional image matching algorithms adopt more complex strategies when dealing with mismatch caused by a lot of noise. In this paper, a simple, intuitive and effective noise processing algorithm is proposed based on Delaunay triangulation in computational geometry. The algorithm extracts feature points using SIFT method, respectively establishes Delaunay triangulation in multi-spectral and panchromatic images, and removes the feature points that three points are collinear and four points are on circle by Delaunay triangulation, obtains the registration images through the correspondence between the Delaunay triangulations. The effect of image registration is evaluated by objective method. In the image matching, Delaunay triangulation is introduced. The establishment of Delaunay triangulation is independent of the selection of initial values. In general, a unique Delaunay triangulation can be got when a feature point set is given, and it can provide the accuracy of the algorithm. The algorithm is simple and clear for converting a lot of mismatch noise to the operation of Delaunay triangulation. Experiment results show the algorithm can keep the good rotation feature and translation invariance in SIFT method, the number of extraction feature points has been significantly reduced in the algorithm compared with SIFT method, registration speed and accuracy are better than the registration algorithm based on conventional SIFT method.