opencv/modules/legacy/src/hmmobs.cpp
2012-10-17 15:57:49 +04:00

635 lines
19 KiB
C++

/*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"
//*F///////////////////////////////////////////////////////////////////////////////////////
// Name: icvImgToObs_DCT_8u32f_C1R
// Purpose: The function takes as input an image and returns the sequnce of observations
// to be used with an embedded HMM; Each observation is top-left block of DCT
// coefficient matrix.
// Context:
// Parameters: img - pointer to the original image ROI
// imgStep - full row width of the image in bytes
// roi - width and height of ROI in pixels
// obs - pointer to resultant observation vectors
// dctSize - size of the block for which DCT is calculated
// obsSize - size of top-left block of DCT coeffs matrix, which is treated
// as observation. Each observation vector consists of
// obsSize.width * obsSize.height floats.
// The following conditions should be satisfied:
// 0 < objSize.width <= dctSize.width,
// 0 < objSize.height <= dctSize.height.
// delta - dctBlocks are overlapped and this parameter specifies horizontal
// and vertical shift.
// Returns:
// CV_NO_ERR or error code
// Notes:
// The algorithm is following:
// 1. First, number of observation vectors per row and per column are calculated:
//
// Nx = floor((roi.width - dctSize.width + delta.width)/delta.width);
// Ny = floor((roi.height - dctSize.height + delta.height)/delta.height);
//
// So, total number of observation vectors is Nx*Ny, and total size of
// array obs must be >= Nx*Ny*obsSize.width*obsSize.height*sizeof(float).
// 2. Observation vectors are calculated in the following loop
// ( actual implementation may be different ), where
// I[x1:x2,y1:y2] means block of pixels from source image with
// x1 <= x < x2, y1 <= y < y2,
// D[x1:x2,y1:y2] means sub matrix of DCT matrix D.
// O[x,y] means observation vector that corresponds to position
// (x*delta.width,y*delta.height) in the source image
// ( all indices are counted from 0 ).
//
// for( y = 0; y < Ny; y++ )
// {
// for( x = 0; x < Nx; x++ )
// {
// D = DCT(I[x*delta.width : x*delta.width + dctSize.width,
// y*delta.height : y*delta.height + dctSize.height]);
// O[x,y] = D[0:obsSize.width, 0:obsSize.height];
// }
// }
//F*/
/*comment out the following line to make DCT be calculated in floating-point arithmetics*/
//#define _CV_INT_DCT
/* for integer DCT only */
#define DCT_SCALE 15
#ifdef _CV_INT_DCT
typedef int work_t;
#define DESCALE CV_DESCALE
#define SCALE(x) CV_FLT_TO_FIX((x),DCT_SCALE)
#else
typedef float work_t;
#define DESCALE(x,n) (float)(x)
#define SCALE(x) (float)(x)
#endif
/* calculate dct transform matrix */
static void icvCalcDCTMatrix( work_t * cfs, int n );
#define MAX_DCT_SIZE 32
static CvStatus CV_STDCALL
icvImgToObs_DCT_8u32f_C1R( uchar * img, int imgStep, CvSize roi,
float *obs, CvSize dctSize,
CvSize obsSize, CvSize delta )
{
/* dct transform matrices: horizontal and vertical */
work_t tab_x[MAX_DCT_SIZE * MAX_DCT_SIZE / 2 + 2];
work_t tab_y[MAX_DCT_SIZE * MAX_DCT_SIZE / 2 + 2];
/* temporary buffers for dct */
work_t temp0[MAX_DCT_SIZE * 4];
work_t temp1[MAX_DCT_SIZE * 4];
work_t *buffer = 0;
work_t *buf_limit;
double s;
int y;
int Nx, Ny;
int n1 = dctSize.height, m1 = n1 / 2;
int n2 = dctSize.width, m2 = n2 / 2;
if( !img || !obs )
return CV_NULLPTR_ERR;
if( roi.width <= 0 || roi.height <= 0 )
return CV_BADSIZE_ERR;
if( delta.width <= 0 || delta.height <= 0 )
return CV_BADRANGE_ERR;
if( obsSize.width <= 0 || dctSize.width < obsSize.width ||
obsSize.height <= 0 || dctSize.height < obsSize.height )
return CV_BADRANGE_ERR;
if( dctSize.width > MAX_DCT_SIZE || dctSize.height > MAX_DCT_SIZE )
return CV_BADRANGE_ERR;
Nx = (roi.width - dctSize.width + delta.width) / delta.width;
Ny = (roi.height - dctSize.height + delta.height) / delta.height;
if( Nx <= 0 || Ny <= 0 )
return CV_BADRANGE_ERR;
buffer = (work_t *)cvAlloc( roi.width * obsSize.height * sizeof( buffer[0] ));
if( !buffer )
return CV_OUTOFMEM_ERR;
icvCalcDCTMatrix( tab_x, dctSize.width );
icvCalcDCTMatrix( tab_y, dctSize.height );
buf_limit = buffer + obsSize.height * roi.width;
for( y = 0; y < Ny; y++, img += delta.height * imgStep )
{
int x, i, j, k;
work_t k0 = 0;
/* do transfroms for each column. Calc only first obsSize.height DCT coefficients */
for( x = 0; x < roi.width; x++ )
{
float is = 0;
work_t *buf = buffer + x;
work_t *tab = tab_y + 2;
if( n1 & 1 )
{
is = img[x + m1 * imgStep];
k0 = ((work_t) is) * tab[-1];
}
/* first coefficient */
for( j = 0; j < m1; j++ )
{
float t0 = img[x + j * imgStep];
float t1 = img[x + (n1 - 1 - j) * imgStep];
float t2 = t0 + t1;
t0 -= t1;
temp0[j] = (work_t) t2;
is += t2;
temp1[j] = (work_t) t0;
}
buf[0] = DESCALE( is * tab[-2], PASS1_SHIFT );
if( (buf += roi.width) >= buf_limit )
continue;
/* other coefficients */
for( ;; )
{
s = 0;
for( k = 0; k < m1; k++ )
s += temp1[k] * tab[k];
buf[0] = DESCALE( s, PASS1_SHIFT );
if( (buf += roi.width) >= buf_limit )
break;
tab += m1;
s = 0;
if( n1 & 1 )
{
k0 = -k0;
s = k0;
}
for( k = 0; k < m1; k++ )
s += temp0[k] * tab[k];
buf[0] = DESCALE( s, PASS1_SHIFT );
tab += m1;
if( (buf += roi.width) >= buf_limit )
break;
}
}
k0 = 0;
/* do transforms for rows. */
for( x = 0; x + dctSize.width <= roi.width; x += delta.width )
{
for( i = 0; i < obsSize.height; i++ )
{
work_t *buf = buffer + x + roi.width * i;
work_t *tab = tab_x + 2;
float *obs_limit = obs + obsSize.width;
s = 0;
if( n2 & 1 )
{
s = buf[m2];
k0 = (work_t) (s * tab[-1]);
}
/* first coefficient */
for( j = 0; j < m2; j++ )
{
work_t t0 = buf[j];
work_t t1 = buf[n2 - 1 - j];
work_t t2 = t0 + t1;
t0 -= t1;
temp0[j] = (work_t) t2;
s += t2;
temp1[j] = (work_t) t0;
}
*obs++ = (float) DESCALE( s * tab[-2], PASS2_SHIFT );
if( obs == obs_limit )
continue;
/* other coefficients */
for( ;; )
{
s = 0;
for( k = 0; k < m2; k++ )
s += temp1[k] * tab[k];
obs[0] = (float) DESCALE( s, PASS2_SHIFT );
if( ++obs == obs_limit )
break;
tab += m2;
s = 0;
if( n2 & 1 )
{
k0 = -k0;
s = k0;
}
for( k = 0; k < m2; k++ )
s += temp0[k] * tab[k];
obs[0] = (float) DESCALE( s, PASS2_SHIFT );
tab += m2;
if( ++obs == obs_limit )
break;
}
}
}
}
cvFree( &buffer );
return CV_NO_ERR;
}
static CvStatus CV_STDCALL
icvImgToObs_DCT_32f_C1R( float * img, int imgStep, CvSize roi,
float *obs, CvSize dctSize,
CvSize obsSize, CvSize delta )
{
/* dct transform matrices: horizontal and vertical */
work_t tab_x[MAX_DCT_SIZE * MAX_DCT_SIZE / 2 + 2];
work_t tab_y[MAX_DCT_SIZE * MAX_DCT_SIZE / 2 + 2];
/* temporary buffers for dct */
work_t temp0[MAX_DCT_SIZE * 4];
work_t temp1[MAX_DCT_SIZE * 4];
work_t *buffer = 0;
work_t *buf_limit;
double s;
int y;
int Nx, Ny;
int n1 = dctSize.height, m1 = n1 / 2;
int n2 = dctSize.width, m2 = n2 / 2;
if( !img || !obs )
return CV_NULLPTR_ERR;
if( roi.width <= 0 || roi.height <= 0 )
return CV_BADSIZE_ERR;
if( delta.width <= 0 || delta.height <= 0 )
return CV_BADRANGE_ERR;
if( obsSize.width <= 0 || dctSize.width < obsSize.width ||
obsSize.height <= 0 || dctSize.height < obsSize.height )
return CV_BADRANGE_ERR;
if( dctSize.width > MAX_DCT_SIZE || dctSize.height > MAX_DCT_SIZE )
return CV_BADRANGE_ERR;
Nx = (roi.width - dctSize.width + delta.width) / delta.width;
Ny = (roi.height - dctSize.height + delta.height) / delta.height;
if( Nx <= 0 || Ny <= 0 )
return CV_BADRANGE_ERR;
buffer = (work_t *)cvAlloc( roi.width * obsSize.height * sizeof( buffer[0] ));
if( !buffer )
return CV_OUTOFMEM_ERR;
icvCalcDCTMatrix( tab_x, dctSize.width );
icvCalcDCTMatrix( tab_y, dctSize.height );
buf_limit = buffer + obsSize.height * roi.width;
imgStep /= sizeof(img[0]);
for( y = 0; y < Ny; y++, img += delta.height * imgStep )
{
int x, i, j, k;
work_t k0 = 0;
/* do transfroms for each column. Calc only first obsSize.height DCT coefficients */
for( x = 0; x < roi.width; x++ )
{
float is = 0;
work_t *buf = buffer + x;
work_t *tab = tab_y + 2;
if( n1 & 1 )
{
is = img[x + m1 * imgStep];
k0 = ((work_t) is) * tab[-1];
}
/* first coefficient */
for( j = 0; j < m1; j++ )
{
float t0 = img[x + j * imgStep];
float t1 = img[x + (n1 - 1 - j) * imgStep];
float t2 = t0 + t1;
t0 -= t1;
temp0[j] = (work_t) t2;
is += t2;
temp1[j] = (work_t) t0;
}
buf[0] = DESCALE( is * tab[-2], PASS1_SHIFT );
if( (buf += roi.width) >= buf_limit )
continue;
/* other coefficients */
for( ;; )
{
s = 0;
for( k = 0; k < m1; k++ )
s += temp1[k] * tab[k];
buf[0] = DESCALE( s, PASS1_SHIFT );
if( (buf += roi.width) >= buf_limit )
break;
tab += m1;
s = 0;
if( n1 & 1 )
{
k0 = -k0;
s = k0;
}
for( k = 0; k < m1; k++ )
s += temp0[k] * tab[k];
buf[0] = DESCALE( s, PASS1_SHIFT );
tab += m1;
if( (buf += roi.width) >= buf_limit )
break;
}
}
k0 = 0;
/* do transforms for rows. */
for( x = 0; x + dctSize.width <= roi.width; x += delta.width )
{
for( i = 0; i < obsSize.height; i++ )
{
work_t *buf = buffer + x + roi.width * i;
work_t *tab = tab_x + 2;
float *obs_limit = obs + obsSize.width;
s = 0;
if( n2 & 1 )
{
s = buf[m2];
k0 = (work_t) (s * tab[-1]);
}
/* first coefficient */
for( j = 0; j < m2; j++ )
{
work_t t0 = buf[j];
work_t t1 = buf[n2 - 1 - j];
work_t t2 = t0 + t1;
t0 -= t1;
temp0[j] = (work_t) t2;
s += t2;
temp1[j] = (work_t) t0;
}
*obs++ = (float) DESCALE( s * tab[-2], PASS2_SHIFT );
if( obs == obs_limit )
continue;
/* other coefficients */
for( ;; )
{
s = 0;
for( k = 0; k < m2; k++ )
s += temp1[k] * tab[k];
obs[0] = (float) DESCALE( s, PASS2_SHIFT );
if( ++obs == obs_limit )
break;
tab += m2;
s = 0;
if( n2 & 1 )
{
k0 = -k0;
s = k0;
}
for( k = 0; k < m2; k++ )
s += temp0[k] * tab[k];
obs[0] = (float) DESCALE( s, PASS2_SHIFT );
tab += m2;
if( ++obs == obs_limit )
break;
}
}
}
}
cvFree( &buffer );
return CV_NO_ERR;
}
static void
icvCalcDCTMatrix( work_t * cfs, int n )
{
static const double sqrt2 = 1.4142135623730950488016887242097;
static const double pi = 3.1415926535897932384626433832795;
static const double sincos[16 * 2] = {
1.00000000000000000, 0.00000000000000006,
0.70710678118654746, 0.70710678118654757,
0.49999999999999994, 0.86602540378443871,
0.38268343236508978, 0.92387953251128674,
0.30901699437494740, 0.95105651629515353,
0.25881904510252074, 0.96592582628906831,
0.22252093395631439, 0.97492791218182362,
0.19509032201612825, 0.98078528040323043,
0.17364817766693033, 0.98480775301220802,
0.15643446504023087, 0.98768834059513777,
0.14231483827328514, 0.98982144188093268,
0.13052619222005157, 0.99144486137381038,
0.12053668025532305, 0.99270887409805397,
0.11196447610330786, 0.99371220989324260,
0.10452846326765346, 0.99452189536827329,
0.09801714032956060, 0.99518472667219693,
};
#define ROTATE( c, s, dc, ds ) \
{ \
t = c*dc - s*ds; \
s = c*ds + s*dc; \
c = t; \
}
#define WRITE2( j, a, b ) \
{ \
cfs[j] = SCALE(a); \
cfs2[j] = SCALE(b); \
}
double t, scale = 1. / sqrt( (double)n );
int i, j, m = n / 2;
cfs[0] = SCALE( scale );
scale *= sqrt2;
cfs[1] = SCALE( scale );
cfs += 2 - m;
if( n > 1 )
{
double a0, b0;
double da0, db0;
work_t *cfs2 = cfs + m * n;
if( n <= 16 )
{
da0 = a0 = sincos[2 * n - 1];
db0 = b0 = sincos[2 * n - 2];
}
else
{
t = pi / (2 * n);
da0 = a0 = cos( t );
db0 = b0 = sin( t );
}
/* other rows */
for( i = 1; i <= m; i++ )
{
double a = a0 * scale;
double b = b0 * scale;
double da = a0 * a0 - b0 * b0;
double db = a0 * b0 + a0 * b0;
cfs += m;
cfs2 -= m;
for( j = 0; j < m; j += 2 )
{
WRITE2( j, a, b );
ROTATE( a, b, da, db );
if( j + 1 < m )
{
WRITE2( j + 1, a, -b );
ROTATE( a, b, da, db );
}
}
ROTATE( a0, b0, da0, db0 );
}
}
#undef ROTATE
#undef WRITE2
}
CV_IMPL void
cvImgToObs_DCT( const void* arr, float *obs, CvSize dctSize,
CvSize obsSize, CvSize delta )
{
CV_FUNCNAME( "cvImgToObs_DCT" );
__BEGIN__;
CvMat stub, *mat = (CvMat*)arr;
CV_CALL( mat = cvGetMat( arr, &stub ));
switch( CV_MAT_TYPE( mat->type ))
{
case CV_8UC1:
IPPI_CALL( icvImgToObs_DCT_8u32f_C1R( mat->data.ptr, mat->step,
cvGetMatSize(mat), obs,
dctSize, obsSize, delta ));
break;
case CV_32FC1:
IPPI_CALL( icvImgToObs_DCT_32f_C1R( mat->data.fl, mat->step,
cvGetMatSize(mat), obs,
dctSize, obsSize, delta ));
break;
default:
CV_ERROR( CV_StsUnsupportedFormat, "" );
}
__END__;
}
/* End of file. */