Image registration using minimum entropic graphs
View Patent ↗In a method of aligning images, a first image and a second image are sampled to a plurality of joint feature pairs. A minimum entropic graph is computed using the joint feature pairs. Using a distance measure, the total length of the minimum entropic graph is measured to obtain a dissimilarity measure. The dissimilarity measure is minimized for updating the geometric transformation.
1. A method of aligning images comprising:
sampling a first image and a second image, obtaining a plurality of joint feature pairs;
computing a minimum entropic graph using the joint feature pairs;
measuring the total length of the minimum entropic graph using a distance measure, obtaining a dissimilarity measure; and
minimizing the dissimilarity measure for updating the geometric transformation;
wherein computing a minimum entropic graph comprises computing a minimum spanning tree; and
wherein computing a minimum spanning tree comprises:
computing the Delaunay triangulation of a point set, obtaining a Delaunay triangulation edge set; and
computing the minimum spanning tree using the Delaunay triangulation edge set and Kruskal's algorithm.
2. A computer readable medium embodying instructions executable by a processor to perform a method of aligning images comprising:
sampling a first image and a second image, obtaining a plurality of joint feature pairs;
computing a minimum entropic graph using the joint feature pairs;
measuring the total length of the minimum entropic graph using a distance measure, obtaining a dissimilarity measure; and
minimizing the dissimilarity measure for updating the geometric transformation;
wherein the joint feature pairs are joint feature pairs in a higher-dimensional feature space, wherein computing a minimum entropic graph comprises computing a minimum spanning tree and wherein computing a minimum spanning tree comprises:
computing the Delaunay triangulation in a reduced-dimensional feature space, wherein the reduced-dimensional feature space is obtained by projection, obtaining a triangulation; and
computing the minimum spanning tree of the triangulation in the higher-dimensional feature space.
3. A computer readable medium including computer code for aligning images executable by a processor to perform a method of aligning images, the computer readable medium comprising:
computer code for sampling a first image and a second image for obtaining a plurality of joint feature pairs;
computer code for computing a minimum entropic graph using the joint feature pairs;
computer code for measuring the total length of the minimum entropic graph using a distance measure for obtaining a dissimilarity measure; and
computer code for minimizing the dissimilarity measure for updating the geometric transformation;
wherein computing a minimum entropic graph comprises computing a minimum spanning tree and
wherein computer code for computing a minimum spanning tree comprises:
computer code for computing the Delaunay triangulation of a point set, obtaining a Delaunay triangulation edge set; and
computer code for computing the minimum spanning tree using the Delaunay triangulation edge set and Kruskal's algorithm.
4. The computer readable medium of claim 3 , wherein the computer code for minimizing the dissimilarity measure for updating the geometric transformation comprises:
computer code for estimating a gradient of the dissimilarity measure with respect to at least one parameter of a geometric transformation; and
computer code for using the gradient of the dissimilarity measure to update the geometric transformation.