Multi-source Image Registration Based On Improved Skeleton Structure Extraction Algorithm

It is possible to acquire all time and high-quality image in complex scene by using multi-source image fusion technology.Image registration is the key of image fusion.Meanwhile,the common representation and extraction of scene features in multi-source images and the best spatial transformation relationship between images are the key points and difficulties of image registration.Structural features are more accurate and robust features in the image.Multi-source image registration based on structural features can achieve better accuracy.The skeleton which is a structural feature is suitable for multi-source image registration because of it is invariance of translation,scaling,rotation and perspective.A skeleton feature extraction algorithm based on Delaunay triangulation is analyzed in detail.In order to improve the performance of existing algorithms under the influence of noise and small deformation,we proposed a method which combined Gaussian pyramid with edge smoothing.With the increasing of the number of pyramid layers,the effect of noise on the object edge in the image decreases.Meanwhile,the further improvement of smoothing of edges is achieved by applying Gaussian filtering on the binary images.Aimed to solve the problem of interference triangle caused by the near distance of edge points in triangulation process,a contour sampling approximation method and threshold filtering strategies are introduced.Firstly,the appropriate sampling interval is determined according to the change of edge gradient,and the threshold value of reserved triangle is calculated.Secondly,the above procedure is repeated to obtain the final set of triangles.The improved method greatly suppresses the generation of interfering triangles.The above improvements are applied to the feature extraction of skeleton structure,and good results are obtained,which lays a foundation for the application of image registration.In this paper,the feature extraction and improvement methods of skeleton structure based on triangulation are completed,and the simulation results and performance comparison analysis of related algorithms are given.Therefore,a full multi-source image registration framework is designed,and simulation experiments are carried out for several typical image registration scenarios.It is proved that the proposed algorithm can extract skeleton structure features accurately and robustly,and can be applied to many kinds of scenes.Image registration experiment shows that it can realize image registration under the condition of low quality multi-source image.