---

RWindowPcaOp Class Reference

#include <RWindowPcaOp.hpp>

Inheritance diagram for RWindowPcaOp:

[legend]Collaboration diagram for RWindowPcaOp:

[legend]List of all members.

---

Detailed Description

PCA on Covariance matrix of Image Windows.

Author:

Bernard De Cuyper

Version:

1.10

Date:

20/12/2003

 
Purpose:        PCA on covariance matrix of color RGB pixels allows to compute a fast separation
                of PCA components.

                Computation of the covariance is reduced, because we are not dealing with large number of channels.
                Also, we have the advantage, that the ordered PCAs can be evaluated quickly.
                The strongest PCA holds the most information. 

Papers: "Principal Component Neural Networks: Theory and Applications", K.I.Diamantaras, S.Y.Kung,
                Wiley-Interscience Publication, 1996.

                "Decision Estimation and Classification: An Introduction to Pattern Recognition and Related Topics.",
                Charles W. Therrien, John Wiley & Sons 1989.

Additional source:
                A very good public routine to tri-diagonalize a real symmetric matrix uses Householder's method.
                Thanks to Jeffrey D. Taft: "The C Eigenvector Source Code Page".

                

@ Copyrights: Bernard De Cuyper & Eddy Fraiha 2004, Eggs & Pictures. MIT/Open BSD copyright model.

Public Methods
	RWindowPcaOp (int widthWindow=8, int heightWindow=8, int pcaChannel=0)
virtual	~RWindowPcaOp ()
virtual RImage *	filter (RImage *source, RImage *destination=0)
	Local full image filtering.
virtual void	report (FILE *file)
virtual void	report ()
Protected Methods
void	computeWindowCovariance (RImage *g)
void	triDiagonalisation ()
void	tri_diag (double *a, double *d, double *e, double *z, int n, double tol)
int	calc_eigenstructure (double *d, double *e, double *z, int n, double macheps)
Private Attributes
int	width
int	height
int	pcaChannel
int	nWindows
int	widthW
int	heightW
int	nwW
int	nhW
int	dimensionW
double *	meanSample
double *	covariance
double *	eigenValue
double *	triDiagonalCov
double *	eigenVectors
double *	e

---

Member Function Documentation

RImage * RWindowPcaOp::filter (  RImage *    source, RImage *    destination = 0 )  [virtual]

Local full image filtering. Parameters: src  is RImage* is RImage source channel dest  is RImage* is RImage result/placeholder Returns : RImage* as result, Implements AbsRImageFilter.

---

The documentation for this class was generated from the following files:

• RWindowPcaOp.hpp
• RWindowPcaOp.cpp

Restoreinpaint sourceforge project `C++/Java Image Processing, Restoration, Inpainting Project'.

Bernard De Cuyper: Open Project Leader: Concept, design and development. Bernard De Cuyper & Eddy Fraiha 2002, 2003. Bernard De Cuyper 2004. Open and free, for friendly usage only. Modifications on Belgium ground of this piece of artistic work, by governement institutions or companies, must be notified to Bernard De Cuyper.