Documentation - Matlab API - MISC - vl_kdtreequery

[INDEX, DIST] = VL_KDTREEQUERY(KDTREE, X, Y) computes the nearest column of X to each column of Y (in Euclidean distance). KDTREE is a forest of kd-trees build by VL_KDTREEBUILD(). X is a NUMDIMENSIONS x NUMDATA data matrix of class SINGLE or DOUBLE with the data indexed by the kd-trees (it must be the same data matrix passed to VK_KDTREEBUILD() to build the trees). Y is the NUMDIMENSIONS x NUMQUERIES matrix of query points and must have the same class of X. INDEX is a 1 x NUMQUERIES matrix of class UINT32 with the index of the nearest column of X for each column of Y. DIST is a 1 x NUMQUERIES vector of class SINGLE or DOUBLE (depending on the class of X and Y) with the corresponding squared Euclidean distances.

[INDEX, DIST] = VL_KDTREEQUERY(..., 'NUMNEIGHBORS', NN) can be used to return the N nearest neighbors rather than just the nearest one. In this case INDEX and DIST are NN x NUMQUERIES matrices. Neighbors are returned by increasing distance.

VL_KDTREEQUERY(..., 'MAXNUMCOMPARISONS', NCOMP) performs at most NCOMP comparisons for each query point. In this case the result is only approximate (i.e. approximated nearest-neighbors, or ANNs) but the speed can be greatly improved.

Options:

NumNeighbors

Sets the number of neighbors to compute for each query point (by default 1).

MaxNumComparisons

Sets the maximum number of comparisons per query point. The special value 0 means unbounded. The default is 0.

See also: VL_KDTREEBUILD(), VL_HELP().