home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!dtix!mimsy!stewart
- From: stewart@cs.umd.edu (G. W. Stewart)
- Newsgroups: sci.math.num-analysis
- Subject: Re: best fit linear transformation
- Message-ID: <63622@mimsy.umd.edu>
- Date: 22 Jan 93 18:13:44 GMT
- References: <NIALL.93Jan18205210@zeste.nmsu.edu> <1993Jan19.113428.9224@ms.uky.edu>
- Sender: news@mimsy.umd.edu
- Distribution: sci
- Organization: U of Maryland, Dept. of Computer Science, Coll. Pk., MD 20742
- Lines: 34
-
- In article <1993Jan19.113428.9224@ms.uky.edu> clarkp@ms.uky.edu (Clark Wells) writes:
- #niall@nmsu.edu (Niall Graham) writes:
- #
- #
- #>The following straight-forward problem has probably been studied
- #>extensively. Any references to it would be greatly appreciated.
- #
- #>Given a pair of matrices X & Y , find a transformation T for which
- #>TX = Y. In general X & Y are inconsistent, so instead of solving
- #>directly for T, we instead seek T which minimizes the sum of the
- #>errors squared in the equation TX-Y = 0.
- #
- #>Thanks in advance,
- #>Niall Graham
- #>NMSU, Las Cruces
- #
- #What you have described is known as the Orthogonal Procustes
- #Problem, and is an application of the Singular Value
- #Decomposition. A good handling of it can be found in _Matrix
- #Computations_ by Gene Golub and Charles van Loan, in their
- #Special Topics section.(I have the 2nd ed.)
- #
- #I hope the reference is meaningful, and I hope it helps.
-
- Actually the Procrustes problem requires that T be orthogonal.
- The present problem can be solved more easily. The problem
- of minimizing the sum of squares of R = TX-Y is equivilent
- to minimizing the sum of squares of the components of each
- row of R. Thus if t_i is the i-th row of T and y_i is the
- i-th row of Y, the sum of squares of the components of
- r_i = t_iX - y_i is to be minimized. This is an ordinary least
- squares problem.
-
- Pete Stewart
-