MKTRAN

Name

mktran -- makes transform matrix from regional weights and eigenvectors.

Synopsis

mktran <regwts_file><eigvecs_file><n_eigvecs_use><tranmat_file><tranmat_file_desc><ascii_outfile>

DESCRIPTION

Mktran takes a matrix of regional weights, and a set of eigenvectors, and makes a transform matrix from the regional weights and the specified number of (first) eigenvectors. The resulting matrix is suitable for transforming an orientation array into a low-dimensional feature vector.

OPTIONS

<regwts_file>

Regional weights file in PCASYS "matrix" format. The dimensions must be 14x15, because that is the pattern of 2x2-vector blocks of orientation vectors. (Usually the output of optrws.)

<eigvecs_file>

Eigenvectors file in PCASYS "matrix" format. The first dimension is the number of eigenvectors contained in the file; the second dimension must be 1680, which is the dimension of an orientation array when it is thought of as a single vector. (Usually the output of eva_evt.)

<n_eigvecs_use>

The number (first) eigenvectors to be used. This will be the first dimension of the resulting transform matrix.

<tranmat_file>

Transform file to be made, in PCASYS "matrix" format. First dimension will be n_eigvecs_use and second dimension will be 1680.

<tranmat_file_desc>

A string to be written into the transform matrix output file as its description string. This string can be of any length, but must not contain embedded newline characters. If it contains spaces, tabs, or shell metacharacters that are not to be expanded, then it should be quoted. To leave the description empty, use '' (two single quotes, i.e. single-quoted empty string). To let mktran make a description, use - (hyphen).

<ascii_outfile>

If y, makes an ascii output file; if n, binary. Binary is recommended, unless the output file must be portable across different byte orders or floating-point formats.

EXAMPLE(S)

From test/pcasys/execs/mktran/mktran.src:

% mktran ../optrws/optrws.bin ../eva_evt/fv1-9.evt 64 fv1-9.opt - n Uses a set of optimized regional weights (optrws.bin) to adjust a eigen-vector basis set (fv1-9.evt) and create a new transformation matrix (fv1-9.opt) that is used to reduce the dimensionality of the feature vectors.

SEE ALSO

eva_evt (1A), lintran (1A), optrws (1A)

AUTHOR

NIST/ITL/DIV894/Image Group