generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

enabled SSE3 by default; integrated SSE3-optimized bilateral filter (by Grigoriy Frolov); modified API of non-local means (use Input/OutputArrays)

Browse Source
This commit is contained in:
Vadim Pisarevsky
2012-08-21 17:16:06 +04:00
parent 9f016da484
commit 2f1cc018c9
5 changed files with 261 additions and 48 deletions
Download Patch File Download Diff File Expand all files Collapse all files

237
modules/imgproc/src/smooth.cpp
View File

@@ -1294,28 +1294,64 @@ class BilateralFilter_8u_Invoker :
public:
BilateralFilter_8u_Invoker(Mat& _dest, const Mat& _temp, int _radius, int _maxk,
int* _space_ofs, float *_space_weight, float *_color_weight) :
ParallelLoopBody(), dest(&_dest), temp(&_temp), radius(_radius),
temp(&_temp), dest(&_dest), radius(_radius),
maxk(_maxk), space_ofs(_space_ofs), space_weight(_space_weight), color_weight(_color_weight)
{
}
virtual void operator() (const Range& range) const
{
int i, j, cn = dest->channels(), k;
Size size = dest->size();
#if CV_SSE3
int CV_DECL_ALIGNED(16) buf[4];
float CV_DECL_ALIGNED(16) bufSum[4];
static const int CV_DECL_ALIGNED(16) bufSignMask[] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
bool haveSSE3 = checkHardwareSupport(CV_CPU_SSE3);
#endif
for( i = range.start; i < range.end; i++ )
{
const uchar* sptr = temp->ptr(i+radius) + radius*cn;
uchar* dptr = dest->ptr(i);
if( cn == 1 )
{
for( j = 0; j < size.width; j++ )
{
float sum = 0, wsum = 0;
int val0 = sptr[j];
for( k = 0; k < maxk; k++ )
k = 0;
#if CV_SSE3
if( haveSSE3 )
{
__m128 _val0 = _mm_set1_ps(val0);
const __m128 _signMask = _mm_load_ps((const float*)bufSignMask);
for( ; k <= maxk - 4; k += 4 )
{
__m128 _valF = _mm_set_ps(sptr[j + space_ofs[k+3]], sptr[j + space_ofs[k+2]],
sptr[j + space_ofs[k+1]], sptr[j + space_ofs[k]]);
__m128 _val = _mm_andnot_ps(_signMask, _mm_sub_ps(_valF, _val0));
_mm_store_si128((__m128i*)buf, _mm_cvtps_epi32(_val));
__m128 _cw = _mm_set_ps(color_weight[buf[3]],color_weight[buf[2]],
color_weight[buf[1]],color_weight[buf[0]]);
__m128 _sw = _mm_loadu_ps(space_weight+k);
__m128 _w = _mm_mul_ps(_cw, _sw);
_cw = _mm_mul_ps(_w, _valF);
_sw = _mm_hadd_ps(_w, _cw);
_sw = _mm_hadd_ps(_sw, _sw);
_mm_storel_pi((__m64*)bufSum, _sw);
sum += bufSum[1];
wsum += bufSum[0];
}
}
#endif
for( ; k < maxk; k++ )
{
int val = sptr[j + space_ofs[k]];
float w = space_weight[k]*color_weight[std::abs(val - val0)];
@@ -1333,7 +1369,57 @@ public:
{
float sum_b = 0, sum_g = 0, sum_r = 0, wsum = 0;
int b0 = sptr[j], g0 = sptr[j+1], r0 = sptr[j+2];
for( k = 0; k < maxk; k++ )
k = 0;
#if CV_SSE3
if( haveSSE3 )
{
const __m128 _b0 = _mm_set1_ps(b0);
const __m128 _g0 = _mm_set1_ps(g0);
const __m128 _r0 = _mm_set1_ps(r0);
const __m128 _signMask = _mm_load_ps((const float*)bufSignMask);
for( ; k <= maxk - 4; k += 4 )
{
const uchar* sptr_k = sptr + j + space_ofs[k];
const uchar* sptr_k1 = sptr + j + space_ofs[k+1];
const uchar* sptr_k2 = sptr + j + space_ofs[k+2];
const uchar* sptr_k3 = sptr + j + space_ofs[k+3];
__m128 _b = _mm_set_ps(sptr_k3[0],sptr_k2[0],sptr_k1[0],sptr_k[0]);
__m128 _g = _mm_set_ps(sptr_k3[1],sptr_k2[1],sptr_k1[1],sptr_k[1]);
__m128 _r = _mm_set_ps(sptr_k3[2],sptr_k2[2],sptr_k1[2],sptr_k[2]);
__m128 bt = _mm_andnot_ps(_signMask, _mm_sub_ps(_b,_b0));
__m128 gt = _mm_andnot_ps(_signMask, _mm_sub_ps(_g,_g0));
__m128 rt = _mm_andnot_ps(_signMask, _mm_sub_ps(_r,_r0));
bt =_mm_add_ps(rt, _mm_add_ps(bt, gt));
_mm_store_si128((__m128i*)buf, _mm_cvtps_epi32(bt));
__m128 _w = _mm_set_ps(color_weight[buf[3]],color_weight[buf[2]],
color_weight[buf[1]],color_weight[buf[0]]);
__m128 _sw = _mm_loadu_ps(space_weight+k);
_w = _mm_mul_ps(_w,_sw);
_b = _mm_mul_ps(_b, _w);
_g = _mm_mul_ps(_g, _w);
_r = _mm_mul_ps(_r, _w);
_w = _mm_hadd_ps(_w, _b);
_g = _mm_hadd_ps(_g, _r);
_w = _mm_hadd_ps(_w, _g);
_mm_store_ps(bufSum, _w);
wsum += bufSum[0];
sum_b += bufSum[1];
sum_g += bufSum[2];
sum_r += bufSum[3];
}
}
#endif
for( ; k < maxk; k++ )
{
const uchar* sptr_k = sptr + j + space_ofs[k];
int b = sptr_k[0], g = sptr_k[1], r = sptr_k[2];
@@ -1351,10 +1437,10 @@ public:
}
}
}
private:
Mat *dest;
const Mat *temp;
Mat *dest;
int radius, maxk, *space_ofs;
float *space_weight, *color_weight;
};
@@ -1364,46 +1450,51 @@ bilateralFilter_8u( const Mat& src, Mat& dst, int d,
double sigma_color, double sigma_space,
int borderType )
{
int cn = src.channels();
int i, j, maxk, radius;
Size size = src.size();
CV_Assert( (src.type() == CV_8UC1 || src.type() == CV_8UC3) &&
src.type() == dst.type() && src.size() == dst.size() &&
src.data != dst.data );
if( sigma_color <= 0 )
sigma_color = 1;
if( sigma_space <= 0 )
sigma_space = 1;
double gauss_color_coeff = -0.5/(sigma_color*sigma_color);
double gauss_space_coeff = -0.5/(sigma_space*sigma_space);
if( d <= 0 )
radius = cvRound(sigma_space*1.5);
else
radius = d/2;
radius = MAX(radius, 1);
d = radius*2 + 1;
Mat temp;
copyMakeBorder( src, temp, radius, radius, radius, radius, borderType );
vector<float> _color_weight(cn*256);
vector<float> _space_weight(d*d);
vector<int> _space_ofs(d*d);
float* color_weight = &_color_weight[0];
float* space_weight = &_space_weight[0];
int* space_ofs = &_space_ofs[0];
// initialize color-related bilateral filter coefficients
for( i = 0; i < 256*cn; i++ )
color_weight[i] = (float)std::exp(i*i*gauss_color_coeff);
// initialize space-related bilateral filter coefficients
for( i = -radius, maxk = 0; i <= radius; i++ )
for( j = -radius; j <= radius; j++ )
{
j = -radius;
for( ;j <= radius; j++ )
{
double r = std::sqrt((double)i*i + (double)j*j);
if( r > radius )
@@ -1411,7 +1502,8 @@ bilateralFilter_8u( const Mat& src, Mat& dst, int d,
space_weight[maxk] = (float)std::exp(r*r*gauss_space_coeff);
space_ofs[maxk++] = (int)(i*temp.step + j*cn);
}
}
BilateralFilter_8u_Invoker body(dst, temp, radius, maxk, space_ofs, space_weight, color_weight);
parallel_for_(Range(0, size.height), body);
}
@@ -1424,7 +1516,7 @@ public:
BilateralFilter_32f_Invoker(int _cn, int _radius, int _maxk, int *_space_ofs,
const Mat& _temp, Mat& _dest, float _scale_index, float *_space_weight, float *_expLUT) :
ParallelLoopBody(), cn(_cn), radius(_radius), maxk(_maxk), space_ofs(_space_ofs),
cn(_cn), radius(_radius), maxk(_maxk), space_ofs(_space_ofs),
temp(&_temp), dest(&_dest), scale_index(_scale_index), space_weight(_space_weight), expLUT(_expLUT)
{
}
@@ -1433,6 +1525,12 @@ public:
{
int i, j, k;
Size size = dest->size();
#if CV_SSE3
int CV_DECL_ALIGNED(16) idxBuf[4];
float CV_DECL_ALIGNED(16) bufSum32[4];
static const int CV_DECL_ALIGNED(16) bufSignMask[] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
bool haveSSE3 = checkHardwareSupport(CV_CPU_SSE3);
#endif
for( i = range.start; i < range.end; i++ )
{
@@ -1445,7 +1543,44 @@ public:
{
float sum = 0, wsum = 0;
float val0 = sptr[j];
for( k = 0; k < maxk; k++ )
k = 0;
#if CV_SSE3
if( haveSSE3 )
{
const __m128 _val0 = _mm_set1_ps(sptr[j]);
const __m128 _scale_index = _mm_set1_ps(scale_index);
const __m128 _signMask = _mm_load_ps((const float*)bufSignMask);
for( ; k <= maxk - 4 ; k += 4 )
{
__m128 _sw = _mm_loadu_ps(space_weight + k);
__m128 _val = _mm_set_ps(sptr[j + space_ofs[k+3]], sptr[j + space_ofs[k+2]],
sptr[j + space_ofs[k+1]], sptr[j + space_ofs[k]]);
__m128 _alpha = _mm_mul_ps(_mm_andnot_ps( _signMask, _mm_sub_ps(_val,_val0)), _scale_index);
__m128i _idx = _mm_cvtps_epi32(_alpha);
_mm_store_si128((__m128i*)idxBuf, _idx);
_alpha = _mm_sub_ps(_alpha, _mm_cvtepi32_ps(_idx));
__m128 _explut = _mm_set_ps(expLUT[idxBuf[3]], expLUT[idxBuf[2]],
expLUT[idxBuf[1]], expLUT[idxBuf[0]]);
__m128 _explut1 = _mm_set_ps(expLUT[idxBuf[3]+1], expLUT[idxBuf[2]+1],
expLUT[idxBuf[1]+1], expLUT[idxBuf[0]+1]);
__m128 _w = _mm_mul_ps(_sw, _mm_add_ps(_explut, _mm_mul_ps(_alpha, _mm_sub_ps(_explut1, _explut))));
_val = _mm_mul_ps(_w, _val);
_sw = _mm_hadd_ps(_w, _val);
_sw = _mm_hadd_ps(_sw, _sw);
_mm_storel_pi((__m64*)bufSum32, _sw);
sum += bufSum32[1];
wsum += bufSum32[0];
}
}
#endif
for( ; k < maxk; k++ )
{
float val = sptr[j + space_ofs[k]];
float alpha = (float)(std::abs(val - val0)*scale_index);
@@ -1465,7 +1600,64 @@ public:
{
float sum_b = 0, sum_g = 0, sum_r = 0, wsum = 0;
float b0 = sptr[j], g0 = sptr[j+1], r0 = sptr[j+2];
for( k = 0; k < maxk; k++ )
k = 0;
#if CV_SSE3
if( haveSSE3 )
{
const __m128 _b0 = _mm_set1_ps(b0);
const __m128 _g0 = _mm_set1_ps(g0);
const __m128 _r0 = _mm_set1_ps(r0);
const __m128 _scale_index = _mm_set1_ps(scale_index);
const __m128 _signMask = _mm_load_ps((const float*)bufSignMask);
for( ; k <= maxk-4; k += 4 )
{
__m128 _sw = _mm_loadu_ps(space_weight + k);
const float* sptr_k = sptr + j + space_ofs[k];
const float* sptr_k1 = sptr + j + space_ofs[k+1];
const float* sptr_k2 = sptr + j + space_ofs[k+2];
const float* sptr_k3 = sptr + j + space_ofs[k+3];
__m128 _b = _mm_set_ps(sptr_k3[0], sptr_k2[0], sptr_k1[0], sptr_k[0]);
__m128 _g = _mm_set_ps(sptr_k3[1], sptr_k2[1], sptr_k1[1], sptr_k[1]);
__m128 _r = _mm_set_ps(sptr_k3[2], sptr_k2[2], sptr_k1[2], sptr_k[2]);
__m128 _bt = _mm_andnot_ps(_signMask,_mm_sub_ps(_b,_b0));
__m128 _gt = _mm_andnot_ps(_signMask,_mm_sub_ps(_g,_g0));
__m128 _rt = _mm_andnot_ps(_signMask,_mm_sub_ps(_r,_r0));
__m128 _alpha = _mm_mul_ps(_scale_index, _mm_add_ps(_rt,_mm_add_ps(_bt, _gt)));
__m128i _idx = _mm_cvtps_epi32(_alpha);
_mm_store_si128((__m128i*)idxBuf, _idx);
_alpha = _mm_sub_ps(_alpha, _mm_cvtepi32_ps(_idx));
__m128 _explut = _mm_set_ps(expLUT[idxBuf[3]], expLUT[idxBuf[2]], expLUT[idxBuf[1]], expLUT[idxBuf[0]]);
__m128 _explut1 = _mm_set_ps(expLUT[idxBuf[3]+1], expLUT[idxBuf[2]+1], expLUT[idxBuf[1]+1], expLUT[idxBuf[0]+1]);
__m128 _w = _mm_mul_ps(_sw, _mm_add_ps(_explut, _mm_mul_ps(_alpha, _mm_sub_ps(_explut1, _explut))));
_b = _mm_mul_ps(_b, _w);
_g = _mm_mul_ps(_g, _w);
_r = _mm_mul_ps(_r, _w);
_w = _mm_hadd_ps(_w, _b);
_g = _mm_hadd_ps(_g, _r);
_w = _mm_hadd_ps(_w, _g);
_mm_store_ps(bufSum32, _w);
wsum += bufSum32[0];
sum_b += bufSum32[1];
sum_g += bufSum32[2];
sum_r += bufSum32[3];
}
}
#endif
for(; k < maxk; k++ )
{
const float* sptr_k = sptr + j + space_ofs[k];
float b = sptr_k[0], g = sptr_k[1], r = sptr_k[2];
@@ -1493,6 +1685,7 @@ private:
Mat *dest;
float scale_index, *space_weight, *expLUT;
};
static void
bilateralFilter_32f( const Mat& src, Mat& dst, int d,
@@ -1569,7 +1762,7 @@ bilateralFilter_32f( const Mat& src, Mat& dst, int d,
}
// initialize space-related bilateral filter coefficients
for( i = -radius, maxk = 0; i <= radius; i++ )
for( i = -radius, maxk = 0; i <= radius; i++ )
for( j = -radius; j <= radius; j++ )
{
double r = std::sqrt((double)i*i + (double)j*j);