mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2024-12-28 03:48:17 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity
02e3afae3a
opencv/modules/legacy/src/eigenobjects.cpp

1812 lines
65 KiB
C++
Raw Normal View History

"atomic bomb" commit. Reorganized OpenCV directory structure
2010-05-12 01:44:00 +08:00
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// Intel License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000, Intel Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of Intel Corporation may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
Set stricter warning rules for gcc
2012-06-08 01:21:29 +08:00
static CvStatus
"atomic bomb" commit. Reorganized OpenCV directory structure
2010-05-12 01:44:00 +08:00
icvJacobiEigens_32f(float *A, float *V, float *E, int n, float eps)
{
int i, j, k, ind;
float *AA = A, *VV = V;
double Amax, anorm = 0, ax;
if( A == NULL || V == NULL || E == NULL )
return CV_NULLPTR_ERR;
if( n <= 0 )
return CV_BADSIZE_ERR;
if( eps < 1.0e-7f )
eps = 1.0e-7f;
/*-------- Prepare --------*/
for( i = 0; i < n; i++, VV += n, AA += n )
{
for( j = 0; j < i; j++ )
{
double Am = AA[j];
anorm += Am * Am;
}
for( j = 0; j < n; j++ )
VV[j] = 0.f;
VV[i] = 1.f;
}
anorm = sqrt( anorm + anorm );
ax = anorm * eps / n;
Amax = anorm;
while( Amax > ax )
{
Amax /= n;
do /* while (ind) */
{
int p, q;
float *V1 = V, *A1 = A;
ind = 0;
for( p = 0; p < n - 1; p++, A1 += n, V1 += n )
{
float *A2 = A + n * (p + 1), *V2 = V + n * (p + 1);
for( q = p + 1; q < n; q++, A2 += n, V2 += n )
{
double x, y, c, s, c2, s2, a;
float *A3, Apq = A1[q], App, Aqq, Aip, Aiq, Vpi, Vqi;
if( fabs( Apq ) < Amax )
continue;
ind = 1;
/*---- Calculation of rotation angle's sine & cosine ----*/
App = A1[p];
Aqq = A2[q];
y = 5.0e-1 * (App - Aqq);
x = -Apq / sqrt( (double)Apq * Apq + (double)y * y );
if( y < 0.0 )
x = -x;
s = x / sqrt( 2.0 * (1.0 + sqrt( 1.0 - (double)x * x )));
s2 = s * s;
c = sqrt( 1.0 - s2 );
c2 = c * c;
a = 2.0 * Apq * c * s;
/*---- Apq annulation ----*/
A3 = A;
for( i = 0; i < p; i++, A3 += n )
{
Aip = A3[p];
Aiq = A3[q];
Vpi = V1[i];
Vqi = V2[i];
A3[p] = (float) (Aip * c - Aiq * s);
A3[q] = (float) (Aiq * c + Aip * s);
V1[i] = (float) (Vpi * c - Vqi * s);
V2[i] = (float) (Vqi * c + Vpi * s);
}
for( ; i < q; i++, A3 += n )
{
Aip = A1[i];
Aiq = A3[q];
Vpi = V1[i];
Vqi = V2[i];
A1[i] = (float) (Aip * c - Aiq * s);
A3[q] = (float) (Aiq * c + Aip * s);
V1[i] = (float) (Vpi * c - Vqi * s);
V2[i] = (float) (Vqi * c + Vpi * s);
}
for( ; i < n; i++ )
{
Aip = A1[i];
Aiq = A2[i];
Vpi = V1[i];
Vqi = V2[i];
A1[i] = (float) (Aip * c - Aiq * s);
A2[i] = (float) (Aiq * c + Aip * s);
V1[i] = (float) (Vpi * c - Vqi * s);
V2[i] = (float) (Vqi * c + Vpi * s);
}
A1[p] = (float) (App * c2 + Aqq * s2 - a);
A2[q] = (float) (App * s2 + Aqq * c2 + a);
A1[q] = A2[p] = 0.0f;
} /*q */
} /*p */
}
while( ind );
Amax /= n;
} /* while ( Amax > ax ) */
for( i = 0, k = 0; i < n; i++, k += n + 1 )
E[i] = A[k];
/*printf(" M = %d\n", M); */
/* -------- ordering -------- */
for( i = 0; i < n; i++ )
{
int m = i;
float Em = (float) fabs( E[i] );
for( j = i + 1; j < n; j++ )
{
float Ej = (float) fabs( E[j] );
m = (Em < Ej) ? j : m;
Em = (Em < Ej) ? Ej : Em;
}
if( m != i )
{
int l;
float b = E[i];
E[i] = E[m];
E[m] = b;
for( j = 0, k = i * n, l = m * n; j < n; j++, k++, l++ )
{
b = V[k];
V[k] = V[l];
V[l] = b;
}
}
}
return CV_NO_ERR;
}
/*F///////////////////////////////////////////////////////////////////////////////////////
// Name: icvCalcCovarMatrixEx_8u32fR
// Purpose: The function calculates a covariance matrix for a group of input objects
// (images, vectors, etc.). ROI supported.
// Context:
// Parameters: nObjects - number of source objects
// objects - array of pointers to ROIs of the source objects
// imgStep - full width of each source object row in bytes
// avg - pointer to averaged object
// avgStep - full width of averaged object row in bytes
// size - ROI size of each source and averaged objects
// covarMatrix - covariance matrix (output parameter; must be allocated
// before call)
//
// Returns: CV_NO_ERR or error code
//
// Notes:
//F*/
static CvStatus CV_STDCALL
icvCalcCovarMatrixEx_8u32fR( int nObjects, void *input, int objStep1,
int ioFlags, int ioBufSize, uchar* buffer,
void *userData, float *avg, int avgStep,
CvSize size, float *covarMatrix )
{
int objStep = objStep1;
/* ---- TEST OF PARAMETERS ---- */
if( nObjects < 2 )
return CV_BADFACTOR_ERR;
if( ioFlags < 0 || ioFlags > 3 )
return CV_BADFACTOR_ERR;
if( ioFlags && ioBufSize < 1024 )
return CV_BADFACTOR_ERR;
if( ioFlags && buffer == NULL )
return CV_NULLPTR_ERR;
if( input == NULL || avg == NULL || covarMatrix == NULL )
return CV_NULLPTR_ERR;
if( size.width > objStep || 4 * size.width > avgStep || size.height < 1 )
return CV_BADSIZE_ERR;
avgStep /= 4;
if( ioFlags & CV_EIGOBJ_INPUT_CALLBACK ) /* ==== USE INPUT CALLBACK ==== */
{
int nio, ngr, igr, n = size.width * size.height, mm = 0;
CvCallback read_callback = ((CvInput *) & input)->callback;
uchar *buffer2;
objStep = n;
nio = ioBufSize / n; /* number of objects in buffer */
ngr = nObjects / nio; /* number of io groups */
if( nObjects % nio )
mm = 1;
ngr += mm;
buffer2 = (uchar *)cvAlloc( sizeof( uchar ) * n );
if( buffer2 == NULL )
return CV_OUTOFMEM_ERR;
for( igr = 0; igr < ngr; igr++ )
{
int k, l;
int io, jo, imin = igr * nio, imax = imin + nio;
uchar *bu1 = buffer, *bu2;
if( imax > nObjects )
imax = nObjects;
/* read igr group */
for( io = imin; io < imax; io++, bu1 += n )
{
CvStatus r;
r = (CvStatus)read_callback( io, (void *) bu1, userData );
if( r )
return r;
}
/* diagonal square calc */
bu1 = buffer;
for( io = imin; io < imax; io++, bu1 += n )
{
bu2 = bu1;
for( jo = io; jo < imax; jo++, bu2 += n )
{
float w = 0.f;
float *fu = avg;
int ij = 0;
for( k = 0; k < size.height; k++, fu += avgStep )
for( l = 0; l < size.width; l++, ij++ )
{
float f = fu[l], u1 = bu1[ij], u2 = bu2[ij];
w += (u1 - f) * (u2 - f);
}
covarMatrix[io * nObjects + jo] = covarMatrix[jo * nObjects + io] = w;
}
}
/* non-diagonal elements calc */
for( jo = imax; jo < nObjects; jo++ )
{
CvStatus r;
bu1 = buffer;
bu2 = buffer2;
/* read jo object */
r = (CvStatus)read_callback( jo, (void *) bu2, userData );
if( r )
return r;
for( io = imin; io < imax; io++, bu1 += n )
{
float w = 0.f;
float *fu = avg;
int ij = 0;
for( k = 0; k < size.height; k++, fu += avgStep )
{
for( l = 0; l < size.width - 3; l += 4, ij += 4 )
{
float f = fu[l];
uchar u1 = bu1[ij];
uchar u2 = bu2[ij];
w += (u1 - f) * (u2 - f);
f = fu[l + 1];
u1 = bu1[ij + 1];
u2 = bu2[ij + 1];
w += (u1 - f) * (u2 - f);
f = fu[l + 2];
u1 = bu1[ij + 2];
u2 = bu2[ij + 2];
w += (u1 - f) * (u2 - f);
f = fu[l + 3];
u1 = bu1[ij + 3];
u2 = bu2[ij + 3];
w += (u1 - f) * (u2 - f);
}
for( ; l < size.width; l++, ij++ )
{
float f = fu[l], u1 = bu1[ij], u2 = bu2[ij];
w += (u1 - f) * (u2 - f);
}
}
covarMatrix[io * nObjects + jo] = covarMatrix[jo * nObjects + io] = w;
}
}
} /* igr */
cvFree( &buffer2 );
} /* if() */
else
/* ==== NOT USE INPUT CALLBACK ==== */
{
int i, j;
uchar **objects = (uchar **) (((CvInput *) & input)->data);
for( i = 0; i < nObjects; i++ )
{
uchar *bu = objects[i];
for( j = i; j < nObjects; j++ )
{
int k, l;
float w = 0.f;
float *a = avg;
uchar *bu1 = bu;
uchar *bu2 = objects[j];
for( k = 0; k < size.height;
k++, bu1 += objStep, bu2 += objStep, a += avgStep )
{
for( l = 0; l < size.width - 3; l += 4 )
{
float f = a[l];
uchar u1 = bu1[l];
uchar u2 = bu2[l];
w += (u1 - f) * (u2 - f);
f = a[l + 1];
u1 = bu1[l + 1];
u2 = bu2[l + 1];
w += (u1 - f) * (u2 - f);
f = a[l + 2];
u1 = bu1[l + 2];
u2 = bu2[l + 2];
w += (u1 - f) * (u2 - f);
f = a[l + 3];
u1 = bu1[l + 3];
u2 = bu2[l + 3];
w += (u1 - f) * (u2 - f);
}
for( ; l < size.width; l++ )
{
float f = a[l];
uchar u1 = bu1[l];
uchar u2 = bu2[l];
w += (u1 - f) * (u2 - f);
}
}
covarMatrix[i * nObjects + j] = covarMatrix[j * nObjects + i] = w;
}
}
} /* else */
return CV_NO_ERR;
}
/*======================== end of icvCalcCovarMatrixEx_8u32fR ===========================*/
static int
icvDefaultBufferSize( void )
{
return 10 * 1024 * 1024;
}
/*F///////////////////////////////////////////////////////////////////////////////////////
// Name: icvCalcEigenObjects_8u32fR
// Purpose: The function calculates an orthonormal eigen basis and a mean (averaged)
// object for a group of input objects (images, vectors, etc.). ROI supported.
// Context:
// Parameters: nObjects - number of source objects
// input - pointer either to array of pointers to input objects
// or to read callback function (depending on ioFlags)
// imgStep - full width of each source object row in bytes
// output - pointer either to array of pointers to output eigen objects
// or to write callback function (depending on ioFlags)
// eigStep - full width of each eigenobject row in bytes
// size - ROI size of each source object
// ioFlags - input/output flags (see Notes)
// ioBufSize - input/output buffer size
// userData - pointer to the structure which contains all necessary
// data for the callback functions
// calcLimit - determines the calculation finish conditions
// avg - pointer to averaged object (has the same size as ROI)
// avgStep - full width of averaged object row in bytes
// eigVals - pointer to corresponding eigenvalues (array of <nObjects>
// elements in descending order)
//
// Returns: CV_NO_ERR or error code
//
// Notes: 1. input/output data (that is, input objects and eigen ones) may either
// be allocated in the RAM or be read from/written to the HDD (or any
// other device) by read/write callback functions. It depends on the
// value of ioFlags paramater, which may be the following:
// CV_EIGOBJ_NO_CALLBACK, or 0;
// CV_EIGOBJ_INPUT_CALLBACK;
// CV_EIGOBJ_OUTPUT_CALLBACK;
// CV_EIGOBJ_BOTH_CALLBACK, or
// CV_EIGOBJ_INPUT_CALLBACK | CV_EIGOBJ_OUTPUT_CALLBACK.
// The callback functions as well as the user data structure must be
// developed by the user.
//
// 2. If ioBufSize = 0, or it's too large, the function dermines buffer size
// itself.
//
// 3. Depending on calcLimit parameter, calculations are finished either if
// eigenfaces number comes up to certain value or the relation of the
// current eigenvalue and the largest one comes down to certain value
// (or any of the above conditions takes place). The calcLimit->type value
// must be CV_TERMCRIT_NUMB, CV_TERMCRIT_EPS or
// CV_TERMCRIT_NUMB | CV_TERMCRIT_EPS. The function returns the real
// values calcLimit->max_iter and calcLimit->epsilon.
//
// 4. eigVals may be equal to NULL (if you don't need eigen values in further).
//
//F*/
static CvStatus CV_STDCALL
icvCalcEigenObjects_8u32fR( int nObjects, void* input, int objStep,
void* output, int eigStep, CvSize size,
int ioFlags, int ioBufSize, void* userData,
CvTermCriteria* calcLimit, float* avg,
int avgStep, float *eigVals )
{
int i, j, n, iev = 0, m1 = nObjects - 1, objStep1 = objStep, eigStep1 = eigStep / 4;
CvSize objSize, eigSize, avgSize;
float *c = 0;
float *ev = 0;
float *bf = 0;
uchar *buf = 0;
void *buffer = 0;
float m = 1.0f / (float) nObjects;
CvStatus r;
if( m1 > calcLimit->max_iter && calcLimit->type != CV_TERMCRIT_EPS )
m1 = calcLimit->max_iter;
/* ---- TEST OF PARAMETERS ---- */
if( nObjects < 2 )
return CV_BADFACTOR_ERR;
if( ioFlags < 0 || ioFlags > 3 )
return CV_BADFACTOR_ERR;
if( input == NULL || output == NULL || avg == NULL )
return CV_NULLPTR_ERR;
if( size.width > objStep || 4 * size.width > eigStep ||
4 * size.width > avgStep || size.height < 1 )
return CV_BADSIZE_ERR;
if( !(ioFlags & CV_EIGOBJ_INPUT_CALLBACK) )
for( i = 0; i < nObjects; i++ )
if( ((uchar **) input)[i] == NULL )
return CV_NULLPTR_ERR;
if( !(ioFlags & CV_EIGOBJ_OUTPUT_CALLBACK) )
for( i = 0; i < m1; i++ )
if( ((float **) output)[i] == NULL )
return CV_NULLPTR_ERR;
avgStep /= 4;
eigStep /= 4;
if( objStep == size.width && eigStep == size.width && avgStep == size.width )
{
size.width *= size.height;
size.height = 1;
objStep = objStep1 = eigStep = eigStep1 = avgStep = size.width;
}
objSize = eigSize = avgSize = size;
if( ioFlags & CV_EIGOBJ_INPUT_CALLBACK )
{
objSize.width *= objSize.height;
objSize.height = 1;
objStep = objSize.width;
objStep1 = size.width;
}
if( ioFlags & CV_EIGOBJ_OUTPUT_CALLBACK )
{
eigSize.width *= eigSize.height;
eigSize.height = 1;
eigStep = eigSize.width;
eigStep1 = size.width;
}
n = objSize.height * objSize.width * (ioFlags & CV_EIGOBJ_INPUT_CALLBACK) +
2 * eigSize.height * eigSize.width * (ioFlags & CV_EIGOBJ_OUTPUT_CALLBACK);
/* Buffer size determination */
if( ioFlags )
{
int size = icvDefaultBufferSize();
ioBufSize = MIN( size, n );
}
/* memory allocation (if necesseay) */
if( ioFlags & CV_EIGOBJ_INPUT_CALLBACK )
{
buf = (uchar *) cvAlloc( sizeof( uchar ) * objSize.width );
if( buf == NULL )
return CV_OUTOFMEM_ERR;
}
if( ioFlags )
{
buffer = (void *) cvAlloc( ioBufSize );
if( buffer == NULL )
{
if( buf )
cvFree( &buf );
return CV_OUTOFMEM_ERR;
}
}
/* Calculation of averaged object */
bf = avg;
for( i = 0; i < avgSize.height; i++, bf += avgStep )
for( j = 0; j < avgSize.width; j++ )
bf[j] = 0.f;
for( i = 0; i < nObjects; i++ )
{
int k, l;
uchar *bu = (ioFlags & CV_EIGOBJ_INPUT_CALLBACK) ? buf : ((uchar **) input)[i];
if( ioFlags & CV_EIGOBJ_INPUT_CALLBACK )
{
CvCallback read_callback = ((CvInput *) & input)->callback;
r = (CvStatus)read_callback( i, (void *) buf, userData );
if( r )
{
if( buffer )
cvFree( &buffer );
if( buf )
cvFree( &buf );
return r;
}
}
bf = avg;
for( k = 0; k < avgSize.height; k++, bf += avgStep, bu += objStep1 )
for( l = 0; l < avgSize.width; l++ )
bf[l] += bu[l];
}
bf = avg;
for( i = 0; i < avgSize.height; i++, bf += avgStep )
for( j = 0; j < avgSize.width; j++ )
bf[j] *= m;
/* Calculation of covariance matrix */
c = (float *) cvAlloc( sizeof( float ) * nObjects * nObjects );
if( c == NULL )
{
if( buffer )
cvFree( &buffer );
if( buf )
cvFree( &buf );
return CV_OUTOFMEM_ERR;
}
r = icvCalcCovarMatrixEx_8u32fR( nObjects, input, objStep1, ioFlags, ioBufSize,
(uchar *) buffer, userData, avg, 4 * avgStep, size, c );
if( r )
{
cvFree( &c );
if( buffer )
cvFree( &buffer );
if( buf )
cvFree( &buf );
return r;
}
/* Calculation of eigenvalues & eigenvectors */
ev = (float *) cvAlloc( sizeof( float ) * nObjects * nObjects );
if( ev == NULL )
{
cvFree( &c );
if( buffer )
cvFree( &buffer );
if( buf )
cvFree( &buf );
return CV_OUTOFMEM_ERR;
}
if( eigVals == NULL )
{
eigVals = (float *) cvAlloc( sizeof( float ) * nObjects );
if( eigVals == NULL )
{
cvFree( &c );
cvFree( &ev );
if( buffer )
cvFree( &buffer );
if( buf )
cvFree( &buf );
return CV_OUTOFMEM_ERR;
}
iev = 1;
}
r = icvJacobiEigens_32f( c, ev, eigVals, nObjects, 0.0f );
cvFree( &c );
if( r )
{
cvFree( &ev );
if( buffer )
cvFree( &buffer );
if( buf )
cvFree( &buf );
if( iev )
cvFree( &eigVals );
return r;
}
/* Eigen objects number determination */
if( calcLimit->type != CV_TERMCRIT_NUMBER )
{
for( i = 0; i < m1; i++ )
if( fabs( eigVals[i] / eigVals[0] ) < calcLimit->epsilon )
break;
m1 = calcLimit->max_iter = i;
}
else
m1 = calcLimit->max_iter;
calcLimit->epsilon = (float) fabs( eigVals[m1 - 1] / eigVals[0] );
for( i = 0; i < m1; i++ )
eigVals[i] = (float) (1.0 / sqrt( (double)eigVals[i] ));
/* ----------------- Calculation of eigenobjects ----------------------- */
if( ioFlags & CV_EIGOBJ_OUTPUT_CALLBACK )
{
int nio, ngr, igr;
nio = ioBufSize / (4 * eigSize.width); /* number of eigen objects in buffer */
ngr = m1 / nio; /* number of io groups */
if( nObjects % nio )
ngr += 1;
for( igr = 0; igr < ngr; igr++ )
{
int i, io, ie, imin = igr * nio, imax = imin + nio;
if( imax > m1 )
imax = m1;
for( i = 0; i < eigSize.width * (imax - imin); i++ )
((float *) buffer)[i] = 0.f;
for( io = 0; io < nObjects; io++ )
{
uchar *bu = ioFlags & CV_EIGOBJ_INPUT_CALLBACK ? buf : ((uchar **) input)[io];
if( ioFlags & CV_EIGOBJ_INPUT_CALLBACK )
{
CvCallback read_callback = ((CvInput *) & input)->callback;
r = (CvStatus)read_callback( io, (void *) buf, userData );
if( r )
{
cvFree( &ev );
if( iev )
cvFree( &eigVals );
if( buffer )
cvFree( &buffer );
if( buf )
cvFree( &buf );
return r;
}
}
for( ie = imin; ie < imax; ie++ )
{
int k, l;
uchar *bv = bu;
float e = ev[ie * nObjects + io] * eigVals[ie];
float *be = ((float *) buffer) + ((ie - imin) * eigStep);
bf = avg;
for( k = 0; k < size.height; k++, bv += objStep1,
bf += avgStep, be += eigStep1 )
{
for( l = 0; l < size.width - 3; l += 4 )
{
float f = bf[l];
uchar v = bv[l];
be[l] += e * (v - f);
f = bf[l + 1];
v = bv[l + 1];
be[l + 1] += e * (v - f);
f = bf[l + 2];
v = bv[l + 2];
be[l + 2] += e * (v - f);
f = bf[l + 3];
v = bv[l + 3];
be[l + 3] += e * (v - f);
}
for( ; l < size.width; l++ )
be[l] += e * (bv[l] - bf[l]);
}
}
} /* io */
for( ie = imin; ie < imax; ie++ ) /* calculated eigen objects writting */
{
CvCallback write_callback = ((CvInput *) & output)->callback;
float *be = ((float *) buffer) + ((ie - imin) * eigStep);
r = (CvStatus)write_callback( ie, (void *) be, userData );
if( r )
{
cvFree( &ev );
if( iev )
cvFree( &eigVals );
if( buffer )
cvFree( &buffer );
if( buf )
cvFree( &buf );
return r;
}
}
} /* igr */
}
else
{
int k, p, l;
for( i = 0; i < m1; i++ ) /* e.o. annulation */
{
float *be = ((float **) output)[i];
for( p = 0; p < eigSize.height; p++, be += eigStep )
for( l = 0; l < eigSize.width; l++ )
be[l] = 0.0f;
}
for( k = 0; k < nObjects; k++ )
{
uchar *bv = (ioFlags & CV_EIGOBJ_INPUT_CALLBACK) ? buf : ((uchar **) input)[k];
if( ioFlags & CV_EIGOBJ_INPUT_CALLBACK )
{
CvCallback read_callback = ((CvInput *) & input)->callback;
r = (CvStatus)read_callback( k, (void *) buf, userData );
if( r )
{
cvFree( &ev );
if( iev )
cvFree( &eigVals );
if( buffer )
cvFree( &buffer );
if( buf )
cvFree( &buf );
return r;
}
}
for( i = 0; i < m1; i++ )
{
float v = eigVals[i] * ev[i * nObjects + k];
float *be = ((float **) output)[i];
uchar *bu = bv;
bf = avg;
for( p = 0; p < size.height; p++, bu += objStep1,
bf += avgStep, be += eigStep1 )
{
for( l = 0; l < size.width - 3; l += 4 )
{
float f = bf[l];
uchar u = bu[l];
be[l] += v * (u - f);
f = bf[l + 1];
u = bu[l + 1];
be[l + 1] += v * (u - f);
f = bf[l + 2];
u = bu[l + 2];
be[l + 2] += v * (u - f);
f = bf[l + 3];
u = bu[l + 3];
be[l + 3] += v * (u - f);
}
for( ; l < size.width; l++ )
be[l] += v * (bu[l] - bf[l]);
}
} /* i */
} /* k */
} /* else */
cvFree( &ev );
if( iev )
cvFree( &eigVals );
else
for( i = 0; i < m1; i++ )
eigVals[i] = 1.f / (eigVals[i] * eigVals[i]);
if( buffer )
cvFree( &buffer );
if( buf )
cvFree( &buf );
return CV_NO_ERR;
}
/* --- End of icvCalcEigenObjects_8u32fR --- */
/*F///////////////////////////////////////////////////////////////////////////////////////
// Name: icvCalcDecompCoeff_8u32fR
// Purpose: The function calculates one decomposition coefficient of input object
// using previously calculated eigen object and the mean (averaged) object
// Context:
// Parameters: obj - input object
// objStep - its step (in bytes)
// eigObj - pointer to eigen object
// eigStep - its step (in bytes)
// avg - pointer to averaged object
// avgStep - its step (in bytes)
// size - ROI size of each source object
//
// Returns: decomposition coefficient value or large negative value (if error)
//
// Notes:
//F*/
static float CV_STDCALL
icvCalcDecompCoeff_8u32fR( uchar* obj, int objStep,
float *eigObj, int eigStep,
float *avg, int avgStep, CvSize size )
{
int i, k;
float w = 0.0f;
if( size.width > objStep || 4 * size.width > eigStep
|| 4 * size.width > avgStep || size.height < 1 )
return -1.0e30f;
if( obj == NULL || eigObj == NULL || avg == NULL )
return -1.0e30f;
eigStep /= 4;
avgStep /= 4;
if( size.width == objStep && size.width == eigStep && size.width == avgStep )
{
size.width *= size.height;
size.height = 1;
objStep = eigStep = avgStep = size.width;
}
for( i = 0; i < size.height; i++, obj += objStep, eigObj += eigStep, avg += avgStep )
{
for( k = 0; k < size.width - 4; k += 4 )
{
float o = (float) obj[k];
float e = eigObj[k];
float a = avg[k];
w += e * (o - a);
o = (float) obj[k + 1];
e = eigObj[k + 1];
a = avg[k + 1];
w += e * (o - a);
o = (float) obj[k + 2];
e = eigObj[k + 2];
a = avg[k + 2];
w += e * (o - a);
o = (float) obj[k + 3];
e = eigObj[k + 3];
a = avg[k + 3];
w += e * (o - a);
}
for( ; k < size.width; k++ )
w += eigObj[k] * ((float) obj[k] - avg[k]);
}
return w;
}
/*F///////////////////////////////////////////////////////////////////////////////////////
// Names: icvEigenDecomposite_8u32fR
// Purpose: The function calculates all decomposition coefficients for input object
// using previously calculated eigen objects basis and the mean (averaged)
// object
// Context:
// Parameters: obj - input object
// objStep - its step (in bytes)
// nEigObjs - number of eigen objects
// eigInput - pointer either to array of pointers to eigen objects
// or to read callback function (depending on ioFlags)
// eigStep - eigen objects step (in bytes)
// ioFlags - input/output flags
// iserData - pointer to the structure which contains all necessary
// data for the callback function
// avg - pointer to averaged object
// avgStep - its step (in bytes)
// size - ROI size of each source object
// coeffs - calculated coefficients (output data)
//
// Returns: icv status
//
// Notes: see notes for icvCalcEigenObjects_8u32fR function
//F*/
static CvStatus CV_STDCALL
icvEigenDecomposite_8u32fR( uchar * obj, int objStep, int nEigObjs,
void *eigInput, int eigStep, int ioFlags,
void *userData, float *avg, int avgStep,
CvSize size, float *coeffs )
{
int i;
if( nEigObjs < 2 )
return CV_BADFACTOR_ERR;
if( ioFlags < 0 || ioFlags > 1 )
return CV_BADFACTOR_ERR;
if( size.width > objStep || 4 * size.width > eigStep ||
4 * size.width > avgStep || size.height < 1 )
return CV_BADSIZE_ERR;
if( obj == NULL || eigInput == NULL || coeffs == NULL || avg == NULL )
return CV_NULLPTR_ERR;
if( !ioFlags )
for( i = 0; i < nEigObjs; i++ )
if( ((uchar **) eigInput)[i] == NULL )
return CV_NULLPTR_ERR;
if( ioFlags ) /* callback */
{
float *buffer;
CvCallback read_callback = ((CvInput *) & eigInput)->callback;
eigStep = 4 * size.width;
/* memory allocation */
buffer = (float *) cvAlloc( sizeof( float ) * size.width * size.height );
if( buffer == NULL )
return CV_OUTOFMEM_ERR;
for( i = 0; i < nEigObjs; i++ )
{
float w;
CvStatus r = (CvStatus)read_callback( i, (void *) buffer, userData );
if( r )
{
cvFree( &buffer );
return r;
}
w = icvCalcDecompCoeff_8u32fR( obj, objStep, buffer,
eigStep, avg, avgStep, size );
if( w < -1.0e29f )
{
cvFree( &buffer );
return CV_NOTDEFINED_ERR;
}
coeffs[i] = w;
}
cvFree( &buffer );
}
else
/* no callback */
for( i = 0; i < nEigObjs; i++ )
{
float w = icvCalcDecompCoeff_8u32fR( obj, objStep, ((float **) eigInput)[i],
eigStep, avg, avgStep, size );
if( w < -1.0e29f )
return CV_NOTDEFINED_ERR;
coeffs[i] = w;
}
return CV_NO_ERR;
}
/*F///////////////////////////////////////////////////////////////////////////////////////
// Names: icvEigenProjection_8u32fR
// Purpose: The function calculates object projection to the eigen sub-space (restores
// an object) using previously calculated eigen objects basis, mean (averaged)
// object and decomposition coefficients of the restored object
// Context:
// Parameters: nEigObjs - Number of eigen objects
// eigens - Array of pointers to eigen objects
// eigStep - Eigen objects step (in bytes)
// coeffs - Previously calculated decomposition coefficients
// avg - Pointer to averaged object
// avgStep - Its step (in bytes)
// rest - Pointer to restored object
// restStep - Its step (in bytes)
// size - ROI size of each object
//
// Returns: CV status
//
// Notes:
//F*/
static CvStatus CV_STDCALL
icvEigenProjection_8u32fR( int nEigObjs, void *eigInput, int eigStep,
int ioFlags, void *userData, float *coeffs,
float *avg, int avgStep, uchar * rest,
int restStep, CvSize size )
{
int i, j, k;
float *buf;
float *buffer = NULL;
float *b;
CvCallback read_callback = ((CvInput *) & eigInput)->callback;
if( size.width > avgStep || 4 * size.width > eigStep || size.height < 1 )
return CV_BADSIZE_ERR;
if( rest == NULL || eigInput == NULL || avg == NULL || coeffs == NULL )
return CV_NULLPTR_ERR;
if( ioFlags < 0 || ioFlags > 1 )
return CV_BADFACTOR_ERR;
if( !ioFlags )
for( i = 0; i < nEigObjs; i++ )
if( ((uchar **) eigInput)[i] == NULL )
return CV_NULLPTR_ERR;
eigStep /= 4;
avgStep /= 4;
if( size.width == restStep && size.width == eigStep && size.width == avgStep )
{
size.width *= size.height;
size.height = 1;
restStep = eigStep = avgStep = size.width;
}
buf = (float *) cvAlloc( sizeof( float ) * size.width * size.height );
if( buf == NULL )
return CV_OUTOFMEM_ERR;
b = buf;
for( i = 0; i < size.height; i++, avg += avgStep, b += size.width )
for( j = 0; j < size.width; j++ )
b[j] = avg[j];
if( ioFlags )
{
buffer = (float *) cvAlloc( sizeof( float ) * size.width * size.height );
if( buffer == NULL )
{
cvFree( &buf );
return CV_OUTOFMEM_ERR;
}
eigStep = size.width;
}
for( k = 0; k < nEigObjs; k++ )
{
float *e = ioFlags ? buffer : ((float **) eigInput)[k];
float c = coeffs[k];
if( ioFlags ) /* read eigen object */
{
CvStatus r = (CvStatus)read_callback( k, (void *) buffer, userData );
if( r )
{
cvFree( &buf );
cvFree( &buffer );
return r;
}
}
b = buf;
for( i = 0; i < size.height; i++, e += eigStep, b += size.width )
{
for( j = 0; j < size.width - 3; j += 4 )
{
float b0 = c * e[j];
float b1 = c * e[j + 1];
float b2 = c * e[j + 2];
float b3 = c * e[j + 3];
b[j] += b0;
b[j + 1] += b1;
b[j + 2] += b2;
b[j + 3] += b3;
}
for( ; j < size.width; j++ )
b[j] += c * e[j];
}
}
b = buf;
for( i = 0; i < size.height; i++, avg += avgStep, b += size.width, rest += restStep )
for( j = 0; j < size.width; j++ )
{
int w = cvRound( b[j] );
w = !(w & ~255) ? w : w < 0 ? 0 : 255;
rest[j] = (uchar) w;
}
cvFree( &buf );
if( ioFlags )
cvFree( &buffer );
return CV_NO_ERR;
}
/*F///////////////////////////////////////////////////////////////////////////////////////
// Name: cvCalcCovarMatrixEx
// Purpose: The function calculates a covariance matrix for a group of input objects
// (images, vectors, etc.).
// Context:
// Parameters: nObjects - number of source objects
// input - pointer either to array of input objects
// or to read callback function (depending on ioFlags)
// ioFlags - input/output flags (see Notes to
// cvCalcEigenObjects function)
// ioBufSize - input/output buffer size
// userData - pointer to the structure which contains all necessary
// data for the callback functions
// avg - averaged object
// covarMatrix - covariance matrix (output parameter; must be allocated
// before call)
//
// Notes: See Notes to cvCalcEigenObjects function
//F*/
CV_IMPL void
cvCalcCovarMatrixEx( int nObjects, void* input, int ioFlags,
int ioBufSize, uchar* buffer, void* userData,
IplImage* avg, float* covarMatrix )
{
float *avg_data;
int avg_step = 0;
CvSize avg_size;
int i;
CV_FUNCNAME( "cvCalcCovarMatrixEx" );
__BEGIN__;
cvGetImageRawData( avg, (uchar **) & avg_data, &avg_step, &avg_size );
if( avg->depth != IPL_DEPTH_32F )
CV_ERROR( CV_BadDepth, cvUnsupportedFormat );
if( avg->nChannels != 1 )
CV_ERROR( CV_BadNumChannels, cvUnsupportedFormat );
if( ioFlags == CV_EIGOBJ_NO_CALLBACK )
{
IplImage **images = (IplImage **) (((CvInput *) & input)->data);
uchar **objects = (uchar **) cvAlloc( sizeof( uchar * ) * nObjects );
int img_step = 0, old_step = 0;
CvSize img_size = avg_size, old_size = avg_size;
if( objects == NULL )
CV_ERROR( CV_StsBadArg, "Insufficient memory" );
for( i = 0; i < nObjects; i++ )
{
IplImage *img = images[i];
uchar *img_data;
cvGetImageRawData( img, &img_data, &img_step, &img_size );
if( img->depth != IPL_DEPTH_8U )
CV_ERROR( CV_BadDepth, cvUnsupportedFormat );
if( img_size != avg_size || img_size != old_size )
CV_ERROR( CV_StsBadArg, "Different sizes of objects" );
if( img->nChannels != 1 )
CV_ERROR( CV_BadNumChannels, cvUnsupportedFormat );
if( i > 0 && img_step != old_step )
CV_ERROR( CV_StsBadArg, "Different steps of objects" );
old_step = img_step;
old_size = img_size;
objects[i] = img_data;
}
CV_CALL( icvCalcCovarMatrixEx_8u32fR( nObjects,
(void*) objects,
img_step,
CV_EIGOBJ_NO_CALLBACK,
0,
NULL,
NULL,
avg_data,
avg_step,
avg_size,
covarMatrix ));
cvFree( &objects );
}
else
{
CV_CALL( icvCalcCovarMatrixEx_8u32fR( nObjects,
input,
avg_step / 4,
ioFlags,
ioBufSize,
buffer,
userData,
avg_data,
avg_step,
avg_size,
covarMatrix ));
}
__END__;
}
/*F///////////////////////////////////////////////////////////////////////////////////////
// Name: cvCalcEigenObjects
// Purpose: The function calculates an orthonormal eigen basis and a mean (averaged)
// object for a group of input objects (images, vectors, etc.).
// Context:
// Parameters: nObjects - number of source objects
// input - pointer either to array of input objects
// or to read callback function (depending on ioFlags)
// output - pointer either to output eigen objects
// or to write callback function (depending on ioFlags)
// ioFlags - input/output flags (see Notes)
// ioBufSize - input/output buffer size
// userData - pointer to the structure which contains all necessary
// data for the callback functions
// calcLimit - determines the calculation finish conditions
// avg - averaged object (has the same size as ROI)
// eigVals - pointer to corresponding eigen values (array of <nObjects>
// elements in descending order)
//
// Notes: 1. input/output data (that is, input objects and eigen ones) may either
// be allocated in the RAM or be read from/written to the HDD (or any
// other device) by read/write callback functions. It depends on the
// value of ioFlags paramater, which may be the following:
// CV_EIGOBJ_NO_CALLBACK, or 0;
// CV_EIGOBJ_INPUT_CALLBACK;
// CV_EIGOBJ_OUTPUT_CALLBACK;
// CV_EIGOBJ_BOTH_CALLBACK, or
// CV_EIGOBJ_INPUT_CALLBACK | CV_EIGOBJ_OUTPUT_CALLBACK.
// The callback functions as well as the user data structure must be
// developed by the user.
//
// 2. If ioBufSize = 0, or it's too large, the function dermines buffer size
// itself.
//
// 3. Depending on calcLimit parameter, calculations are finished either if
// eigenfaces number comes up to certain value or the relation of the
// current eigenvalue and the largest one comes down to certain value
// (or any of the above conditions takes place). The calcLimit->type value
// must be CV_TERMCRIT_NUMB, CV_TERMCRIT_EPS or
// CV_TERMCRIT_NUMB | CV_TERMCRIT_EPS. The function returns the real
// values calcLimit->max_iter and calcLimit->epsilon.
//
// 4. eigVals may be equal to NULL (if you don't need eigen values in further).
//
//F*/
CV_IMPL void
cvCalcEigenObjects( int nObjects,
void* input,
void* output,
int ioFlags,
int ioBufSize,
void* userData,
CvTermCriteria* calcLimit,
IplImage* avg,
float* eigVals )
{
float *avg_data;
int avg_step = 0;
CvSize avg_size;
int i;
int nEigens = nObjects - 1;
CV_FUNCNAME( "cvCalcEigenObjects" );
__BEGIN__;
cvGetImageRawData( avg, (uchar **) & avg_data, &avg_step, &avg_size );
if( avg->depth != IPL_DEPTH_32F )
CV_ERROR( CV_BadDepth, cvUnsupportedFormat );
if( avg->nChannels != 1 )
CV_ERROR( CV_BadNumChannels, cvUnsupportedFormat );
if( nEigens > calcLimit->max_iter && calcLimit->type != CV_TERMCRIT_EPS )
nEigens = calcLimit->max_iter;
switch (ioFlags)
{
case CV_EIGOBJ_NO_CALLBACK:
{
IplImage **objects = (IplImage **) (((CvInput *) & input)->data);
IplImage **eigens = (IplImage **) (((CvInput *) & output)->data);
uchar **objs = (uchar **) cvAlloc( sizeof( uchar * ) * nObjects );
float **eigs = (float **) cvAlloc( sizeof( float * ) * nEigens );
int obj_step = 0, old_step = 0;
int eig_step = 0, oldeig_step = 0;
CvSize obj_size = avg_size, old_size = avg_size,
eig_size = avg_size, oldeig_size = avg_size;
if( objects == NULL || eigens == NULL )
CV_ERROR( CV_StsBadArg, "Insufficient memory" );
for( i = 0; i < nObjects; i++ )
{
IplImage *img = objects[i];
uchar *obj_data;
cvGetImageRawData( img, &obj_data, &obj_step, &obj_size );
if( img->depth != IPL_DEPTH_8U )
CV_ERROR( CV_BadDepth, cvUnsupportedFormat );
if( obj_size != avg_size || obj_size != old_size )
CV_ERROR( CV_StsBadArg, "Different sizes of objects" );
if( img->nChannels != 1 )
CV_ERROR( CV_BadNumChannels, cvUnsupportedFormat );
if( i > 0 && obj_step != old_step )
CV_ERROR( CV_StsBadArg, "Different steps of objects" );
old_step = obj_step;
old_size = obj_size;
objs[i] = obj_data;
}
for( i = 0; i < nEigens; i++ )
{
IplImage *eig = eigens[i];
float *eig_data;
cvGetImageRawData( eig, (uchar **) & eig_data, &eig_step, &eig_size );
if( eig->depth != IPL_DEPTH_32F )
CV_ERROR( CV_BadDepth, cvUnsupportedFormat );
if( eig_size != avg_size || eig_size != oldeig_size )
CV_ERROR( CV_StsBadArg, "Different sizes of objects" );
if( eig->nChannels != 1 )
CV_ERROR( CV_BadNumChannels, cvUnsupportedFormat );
if( i > 0 && eig_step != oldeig_step )
CV_ERROR( CV_StsBadArg, "Different steps of objects" );
oldeig_step = eig_step;
oldeig_size = eig_size;
eigs[i] = eig_data;
}
CV_CALL( icvCalcEigenObjects_8u32fR( nObjects, (void*) objs, obj_step,
(void*) eigs, eig_step, obj_size,
ioFlags, ioBufSize, userData,
calcLimit, avg_data, avg_step, eigVals ));
cvFree( &objs );
cvFree( &eigs );
break;
}
case CV_EIGOBJ_OUTPUT_CALLBACK:
{
IplImage **objects = (IplImage **) (((CvInput *) & input)->data);
uchar **objs = (uchar **) cvAlloc( sizeof( uchar * ) * nObjects );
int obj_step = 0, old_step = 0;
CvSize obj_size = avg_size, old_size = avg_size;
if( objects == NULL )
CV_ERROR( CV_StsBadArg, "Insufficient memory" );
for( i = 0; i < nObjects; i++ )
{
IplImage *img = objects[i];
uchar *obj_data;
cvGetImageRawData( img, &obj_data, &obj_step, &obj_size );
if( img->depth != IPL_DEPTH_8U )
CV_ERROR( CV_BadDepth, cvUnsupportedFormat );
if( obj_size != avg_size || obj_size != old_size )
CV_ERROR( CV_StsBadArg, "Different sizes of objects" );
if( img->nChannels != 1 )
CV_ERROR( CV_BadNumChannels, cvUnsupportedFormat );
if( i > 0 && obj_step != old_step )
CV_ERROR( CV_StsBadArg, "Different steps of objects" );
old_step = obj_step;
old_size = obj_size;
objs[i] = obj_data;
}
CV_CALL( icvCalcEigenObjects_8u32fR( nObjects,
(void*) objs,
obj_step,
output,
avg_step,
obj_size,
ioFlags,
ioBufSize,
userData,
calcLimit,
avg_data,
avg_step,
eigVals ));
cvFree( &objs );
break;
}
case CV_EIGOBJ_INPUT_CALLBACK:
{
IplImage **eigens = (IplImage **) (((CvInput *) & output)->data);
float **eigs = (float**) cvAlloc( sizeof( float* ) * nEigens );
int eig_step = 0, oldeig_step = 0;
CvSize eig_size = avg_size, oldeig_size = avg_size;
if( eigens == NULL )
CV_ERROR( CV_StsBadArg, "Insufficient memory" );
for( i = 0; i < nEigens; i++ )
{
IplImage *eig = eigens[i];
float *eig_data;
cvGetImageRawData( eig, (uchar **) & eig_data, &eig_step, &eig_size );
if( eig->depth != IPL_DEPTH_32F )
CV_ERROR( CV_BadDepth, cvUnsupportedFormat );
if( eig_size != avg_size || eig_size != oldeig_size )
CV_ERROR( CV_StsBadArg, "Different sizes of objects" );
if( eig->nChannels != 1 )
CV_ERROR( CV_BadNumChannels, cvUnsupportedFormat );
if( i > 0 && eig_step != oldeig_step )
CV_ERROR( CV_StsBadArg, "Different steps of objects" );
oldeig_step = eig_step;
oldeig_size = eig_size;
eigs[i] = eig_data;
}
CV_CALL( icvCalcEigenObjects_8u32fR( nObjects,
input,
avg_step / 4,
(void*) eigs,
eig_step,
eig_size,
ioFlags,
ioBufSize,
userData,
calcLimit,
avg_data,
avg_step,
eigVals ));
cvFree( &eigs );
break;
}
case CV_EIGOBJ_INPUT_CALLBACK | CV_EIGOBJ_OUTPUT_CALLBACK:
CV_CALL( icvCalcEigenObjects_8u32fR( nObjects,
input,
avg_step / 4,
output,
avg_step,
avg_size,
ioFlags,
ioBufSize,
userData,
calcLimit,
avg_data,
avg_step,
eigVals ));
break;
default:
CV_ERROR( CV_StsBadArg, "Unsupported i/o flag" );
}
__END__;
}
/*--------------------------------------------------------------------------------------*/
/*F///////////////////////////////////////////////////////////////////////////////////////
// Name: cvCalcDecompCoeff
// Purpose: The function calculates one decomposition coefficient of input object
// using previously calculated eigen object and the mean (averaged) object
// Context:
// Parameters: obj - input object
// eigObj - eigen object
// avg - averaged object
//
// Returns: decomposition coefficient value or large negative value (if error)
//
// Notes:
//F*/
CV_IMPL double
cvCalcDecompCoeff( IplImage * obj, IplImage * eigObj, IplImage * avg )
{
double coeff = DBL_MAX;
uchar *obj_data;
float *eig_data;
float *avg_data;
int obj_step = 0, eig_step = 0, avg_step = 0;
CvSize obj_size, eig_size, avg_size;
CV_FUNCNAME( "cvCalcDecompCoeff" );
__BEGIN__;
cvGetImageRawData( obj, &obj_data, &obj_step, &obj_size );
if( obj->depth != IPL_DEPTH_8U )
CV_ERROR( CV_BadDepth, cvUnsupportedFormat );
if( obj->nChannels != 1 )
CV_ERROR( CV_BadNumChannels, cvUnsupportedFormat );
cvGetImageRawData( eigObj, (uchar **) & eig_data, &eig_step, &eig_size );
if( eigObj->depth != IPL_DEPTH_32F )
CV_ERROR( CV_BadDepth, cvUnsupportedFormat );
if( eigObj->nChannels != 1 )
CV_ERROR( CV_BadNumChannels, cvUnsupportedFormat );
cvGetImageRawData( avg, (uchar **) & avg_data, &avg_step, &avg_size );
if( avg->depth != IPL_DEPTH_32F )
CV_ERROR( CV_BadDepth, cvUnsupportedFormat );
if( avg->nChannels != 1 )
CV_ERROR( CV_BadNumChannels, cvUnsupportedFormat );
if( obj_size != eig_size || obj_size != avg_size )
CV_ERROR( CV_StsBadArg, "different sizes of images" );
coeff = icvCalcDecompCoeff_8u32fR( obj_data, obj_step,
eig_data, eig_step,
avg_data, avg_step, obj_size );
__END__;
return coeff;
}
/*--------------------------------------------------------------------------------------*/
/*F///////////////////////////////////////////////////////////////////////////////////////
// Names: cvEigenDecomposite
// Purpose: The function calculates all decomposition coefficients for input object
// using previously calculated eigen objects basis and the mean (averaged)
// object
//
// Parameters: obj - input object
// nEigObjs - number of eigen objects
// eigInput - pointer either to array of pointers to eigen objects
// or to read callback function (depending on ioFlags)
// ioFlags - input/output flags
// userData - pointer to the structure which contains all necessary
// data for the callback function
// avg - averaged object
// coeffs - calculated coefficients (output data)
//
// Notes: see notes for cvCalcEigenObjects function
//F*/
CV_IMPL void
cvEigenDecomposite( IplImage* obj,
int nEigObjs,
void* eigInput,
int ioFlags,
void* userData,
IplImage* avg,
float* coeffs )
{
float *avg_data;
uchar *obj_data;
int avg_step = 0, obj_step = 0;
CvSize avg_size, obj_size;
int i;
CV_FUNCNAME( "cvEigenDecomposite" );
__BEGIN__;
cvGetImageRawData( avg, (uchar **) & avg_data, &avg_step, &avg_size );
if( avg->depth != IPL_DEPTH_32F )
CV_ERROR( CV_BadDepth, cvUnsupportedFormat );
if( avg->nChannels != 1 )
CV_ERROR( CV_BadNumChannels, cvUnsupportedFormat );
cvGetImageRawData( obj, &obj_data, &obj_step, &obj_size );
if( obj->depth != IPL_DEPTH_8U )
CV_ERROR( CV_BadDepth, cvUnsupportedFormat );
if( obj->nChannels != 1 )
CV_ERROR( CV_BadNumChannels, cvUnsupportedFormat );
if( obj_size != avg_size )
CV_ERROR( CV_StsBadArg, "Different sizes of objects" );
if( ioFlags == CV_EIGOBJ_NO_CALLBACK )
{
IplImage **eigens = (IplImage **) (((CvInput *) & eigInput)->data);
float **eigs = (float **) cvAlloc( sizeof( float * ) * nEigObjs );
int eig_step = 0, old_step = 0;
CvSize eig_size = avg_size, old_size = avg_size;
if( eigs == NULL )
CV_ERROR( CV_StsBadArg, "Insufficient memory" );
for( i = 0; i < nEigObjs; i++ )
{
IplImage *eig = eigens[i];
float *eig_data;
cvGetImageRawData( eig, (uchar **) & eig_data, &eig_step, &eig_size );
if( eig->depth != IPL_DEPTH_32F )
CV_ERROR( CV_BadDepth, cvUnsupportedFormat );
if( eig_size != avg_size || eig_size != old_size )
CV_ERROR( CV_StsBadArg, "Different sizes of objects" );
if( eig->nChannels != 1 )
CV_ERROR( CV_BadNumChannels, cvUnsupportedFormat );
if( i > 0 && eig_step != old_step )
CV_ERROR( CV_StsBadArg, "Different steps of objects" );
old_step = eig_step;
old_size = eig_size;
eigs[i] = eig_data;
}
CV_CALL( icvEigenDecomposite_8u32fR( obj_data,
obj_step,
nEigObjs,
(void*) eigs,
eig_step,
ioFlags,
userData,
avg_data,
avg_step,
obj_size,
coeffs ));
cvFree( &eigs );
}
else
{
CV_CALL( icvEigenDecomposite_8u32fR( obj_data,
obj_step,
nEigObjs,
eigInput,
avg_step,
ioFlags,
userData,
avg_data,
avg_step,
obj_size,
coeffs ));
}
__END__;
}
/*--------------------------------------------------------------------------------------*/
/*F///////////////////////////////////////////////////////////////////////////////////////
// Name: cvEigenProjection
// Purpose: The function calculates object projection to the eigen sub-space (restores
// an object) using previously calculated eigen objects basis, mean (averaged)
// object and decomposition coefficients of the restored object
// Context:
// Parameters: nEigObjs - number of eigen objects
// eigInput - pointer either to array of pointers to eigen objects
// or to read callback function (depending on ioFlags)
// ioFlags - input/output flags
// userData - pointer to the structure which contains all necessary
// data for the callback function
// coeffs - array of decomposition coefficients
// avg - averaged object
// proj - object projection (output data)
//
// Notes: see notes for cvCalcEigenObjects function
//F*/
CV_IMPL void
cvEigenProjection( void* eigInput,
int nEigObjs,
int ioFlags,
void* userData,
float* coeffs,
IplImage* avg,
IplImage* proj )
{
float *avg_data;
uchar *proj_data;
int avg_step = 0, proj_step = 0;
CvSize avg_size, proj_size;
int i;
CV_FUNCNAME( "cvEigenProjection" );
__BEGIN__;
cvGetImageRawData( avg, (uchar **) & avg_data, &avg_step, &avg_size );
if( avg->depth != IPL_DEPTH_32F )
CV_ERROR( CV_BadDepth, cvUnsupportedFormat );
if( avg->nChannels != 1 )
CV_ERROR( CV_BadNumChannels, cvUnsupportedFormat );
cvGetImageRawData( proj, &proj_data, &proj_step, &proj_size );
if( proj->depth != IPL_DEPTH_8U )
CV_ERROR( CV_BadDepth, cvUnsupportedFormat );
if( proj->nChannels != 1 )
CV_ERROR( CV_BadNumChannels, cvUnsupportedFormat );
if( proj_size != avg_size )
CV_ERROR( CV_StsBadArg, "Different sizes of projects" );
if( ioFlags == CV_EIGOBJ_NO_CALLBACK )
{
IplImage **eigens = (IplImage**) (((CvInput *) & eigInput)->data);
float **eigs = (float**) cvAlloc( sizeof( float * ) * nEigObjs );
int eig_step = 0, old_step = 0;
CvSize eig_size = avg_size, old_size = avg_size;
if( eigs == NULL )
CV_ERROR( CV_StsBadArg, "Insufficient memory" );
for( i = 0; i < nEigObjs; i++ )
{
IplImage *eig = eigens[i];
float *eig_data;
cvGetImageRawData( eig, (uchar **) & eig_data, &eig_step, &eig_size );
if( eig->depth != IPL_DEPTH_32F )
CV_ERROR( CV_BadDepth, cvUnsupportedFormat );
if( eig_size != avg_size || eig_size != old_size )
CV_ERROR( CV_StsBadArg, "Different sizes of objects" );
if( eig->nChannels != 1 )
CV_ERROR( CV_BadNumChannels, cvUnsupportedFormat );
if( i > 0 && eig_step != old_step )
CV_ERROR( CV_StsBadArg, "Different steps of objects" );
old_step = eig_step;
old_size = eig_size;
eigs[i] = eig_data;
}
CV_CALL( icvEigenProjection_8u32fR( nEigObjs,
(void*) eigs,
eig_step,
ioFlags,
userData,
coeffs,
avg_data,
avg_step,
proj_data,
proj_step,
avg_size ));
cvFree( &eigs );
}
else
{
CV_CALL( icvEigenProjection_8u32fR( nEigObjs,
eigInput,
avg_step,
ioFlags,
userData,
coeffs,
avg_data,
avg_step,
proj_data,
proj_step,
avg_size ));
}
__END__;
}
/* End of file. */
Reference in New Issue Copy Permalink