Table of Contents  ▸  List of Commands  ▸  Matrix Computation  ▸  mproj◀  orthogonalize    |    solve  ▶


Built-in command



Find best matching projection of selected matrices onto the span of an over-complete

dictionary D, using the orthogonal projection or Matching Pursuit algorithm.
Selected images are 2D-matrices in which each column represent a signal to project.
[dictionary] is a matrix in which each column is an element of the dictionary D.
method tells what projection algorithm must be applied. It can be:
- 0 = orthogonal projection (least-squares solution using LU-based solver).
- 1 = matching pursuit.
- 2 = matching pursuit, with a single orthogonal projection step at the end.
- >=3 = orthogonal matching pursuit where an orthogonal projection step is performed
every method-2 iterations.
max_iter sets the max number of iterations processed for each signal.
If set to 0 (default), max_iter is equal to the number of columns in D.
(only meaningful for matching pursuit and its variants).
max_residual gives a stopping criterion on signal reconstruction accuracy.
(only meaningful for matching pursuit and its variants).
For each selected image, the result is returned as a matrix W
whose columns correspond to the weights associated to each column of D,
such that the matrix product D*W is an approximation of the input matrix.

Default values:

method=0, max_iter=0 and max_residual=1e-6.