static void EvaluateNoise( FVector& noiseEstimates,
FVector& noiseFractions,
StringList& noiseAlgorithms,
const Image& image, int algorithm )
{
SpinStatus spin;
image.SetStatusCallback( &spin );
image.Status().Initialize( "Noise evaluation" );
image.Status().DisableInitialization();
Console console;
int numberOfThreads = Thread::NumberOfThreads( image.NumberOfPixels(), 1 );
if ( numberOfThreads >= 3 )
{
int numberOfSubthreads = RoundI( numberOfThreads/3.0 );
ReferenceArray<NoiseEvaluationThread> threads;
threads.Add( new NoiseEvaluationThread( image, 0, algorithm, numberOfSubthreads ) );
threads.Add( new NoiseEvaluationThread( image, 1, algorithm, numberOfSubthreads ) );
threads.Add( new NoiseEvaluationThread( image, 2, algorithm, numberOfThreads - 2*numberOfSubthreads ) );
AbstractImage::ThreadData data( image, 0 ); // unbounded
AbstractImage::RunThreads( threads, data );
for ( int i = 0; i < 3; ++i )
{
noiseEstimates[i] = threads[i].noiseEstimate;
noiseFractions[i] = threads[i].noiseFraction;
noiseAlgorithms[i] = String( threads[i].noiseAlgorithm );
}
threads.Destroy();
}
else
{
for ( int i = 0; i < 3; ++i )
{
NoiseEvaluationThread thread( image, i, algorithm, 1 );
thread.Run();
noiseEstimates[i] = thread.noiseEstimate;
noiseFractions[i] = thread.noiseFraction;
noiseAlgorithms[i] = String( thread.noiseAlgorithm );
}
}
image.ResetSelections();
image.Status().Complete();
console.WriteLn( "<end><cbr>Gaussian noise estimates:" );
for ( int i = 0; i < 3; ++i )
console.WriteLn( String().Format( "s%d = %.3e, n%d = %.4f ",
i, noiseEstimates[i], i, noiseFractions[i] ) + '(' + noiseAlgorithms[i] + ')' );
}
static void SuperPixelThreaded( Image& target, const GenericImage<P>& source, const DebayerInstance& instance )
{
int target_w = source.Width() >> 1;
int target_h = source.Height() >> 1;
target.AllocateData( target_w, target_h, 3, ColorSpace::RGB );
target.Status().Initialize( "SuperPixel debayering", target_h );
int numberOfThreads = Thread::NumberOfThreads( target_h, 1 );
int rowsPerThread = target_h/numberOfThreads;
AbstractImage::ThreadData data( target, target_h );
ReferenceArray<SuperPixelThread<P> > threads;
for ( int i = 0, j = 1; i < numberOfThreads; ++i, ++j )
threads.Add( new SuperPixelThread<P>( data, target, source, instance,
i*rowsPerThread,
(j < numberOfThreads) ? j*rowsPerThread : target_h ) );
AbstractImage::RunThreads( threads, data );
threads.Destroy();
target.Status() = data.status;
}
static void VNGThreaded( Image& target, const GenericImage<P>& source, const DebayerInstance& instance )
{
int target_w = source.Width();
int target_h = source.Height();
target.AllocateData( target_w, target_h, 3, ColorSpace::RGB );
target.Status().Initialize( "VNG debayering", target_h-4 );
int numberOfThreads = Thread::NumberOfThreads( target_h-4, 1 );
int rowsPerThread = (target_h - 4)/numberOfThreads;
AbstractImage::ThreadData data( target, target_h-4 );
ReferenceArray<VNGThread<P> > threads;
for ( int i = 0, j = 1; i < numberOfThreads; ++i, ++j )
threads.Add( new VNGThread<P>( data, target, source, instance,
i*rowsPerThread + 2,
(j < numberOfThreads) ? j*rowsPerThread + 2 : target_h-2 ) );
AbstractImage::RunThreads( threads, data );
threads.Destroy();
// copy top and bottom two rows from the adjecent ones
for ( int col = 0; col < target_w; col++ )
for ( int i = 0; i < 3; i++ )
{
target.Pixel( col, 0, i ) = target.Pixel( col, 1, i ) = target.Pixel( col, 2, i );
target.Pixel( col, target_h-1, i ) = target.Pixel( col, target_h-2, i ) = target.Pixel( col, target_h-3, i );
}
target.Status() = data.status;
}
void HistogramTransformation::Make24BitLUT( uint16* lut ) const
{
if ( lut == 0 )
return;
int numberOfThreads = m_parallel ? Min( int( m_maxProcessors ), Thread::NumberOfThreads( uint24_max+1, 256 ) ) : 1;
int itemsPerThread = (uint24_max + 1)/numberOfThreads;
bool useAffinity = m_parallel && Thread::IsRootThread();
ReferenceArray<LUT2416Thread> threads;
for ( int i = 0, j = 1; i < numberOfThreads; ++i, ++j )
threads.Add( new LUT2416Thread( lut, *this,
i*itemsPerThread, (j < numberOfThreads) ? j*itemsPerThread : uint24_max+1 ) );
if ( numberOfThreads > 1 )
{
for ( int i = 0; i < numberOfThreads; ++i )
threads[i].Start( ThreadPriority::DefaultMax, useAffinity ? i : -1 );
for ( int i = 0; i < numberOfThreads; ++i )
threads[i].Wait();
}
else
threads[0].Run();
threads.Destroy();
}
template <class P> static
void Apply( GenericImage<P>& image, const HistogramTransformation& H )
{
if ( image.IsEmptySelection() )
return;
image.EnsureUnique();
Rect r = image.SelectedRectangle();
int h = r.Height();
int numberOfThreads = H.IsParallelProcessingEnabled() ? Min( H.MaxProcessors(), pcl::Thread::NumberOfThreads( h, 1 ) ) : 1;
int rowsPerThread = h/numberOfThreads;
size_type N = image.NumberOfSelectedSamples();
if ( image.Status().IsInitializationEnabled() )
image.Status().Initialize( "Histogram transformation", N );
ThreadData<P> data( image, H, N );
ReferenceArray<Thread<P> > threads;
for ( int i = 0, j = 1; i < numberOfThreads; ++i, ++j )
threads.Add( new Thread<P>( data, i*rowsPerThread, (j < numberOfThreads) ? j*rowsPerThread : h ) );
AbstractImage::RunThreads( threads, data );
threads.Destroy();
image.Status() = data.status;
}
static void Apply( GenericImage<P>& image, const ColorSaturationInstance& instance, bool useLUT = false )
{
if ( instance.Curve().IsIdentity() )
{
Console().WriteLn( "<end><cbr><* Identity *>" );
return;
}
size_type N = image.NumberOfPixels();
int numberOfThreads = Thread::NumberOfThreads( N, 16 );
size_type pixelsPerThread = N/numberOfThreads;
image.Status().Initialize( "Color saturation transformation, HSVL space", N );
ThreadData data( image, N );
if ( useLUT )
data.lut = MakeLUT( instance );
ReferenceArray<ColorSaturationThread<P> > threads;
for ( int i = 0, j = 1; i < numberOfThreads; ++i, ++j )
threads.Add( new ColorSaturationThread<P>( instance, data, image,
i*pixelsPerThread,
(j < numberOfThreads) ? j*pixelsPerThread : N ) );
AbstractImage::RunThreads( threads, data );
threads.Destroy();
image.Status() = data.status;
}
static void Apply( GenericImage<P>& image, const CurvesTransformationInstance& instance, bool useLUT = false )
{
int numberOfCurves = 0;
if ( !instance[CurveIndex::RGBK].IsIdentity() )
numberOfCurves = image.NumberOfNominalChannels();
if ( image.IsColor() )
{
for ( int c = 0; c < image.NumberOfNominalChannels(); ++c )
if ( !instance[c].IsIdentity() )
++numberOfCurves;
if ( !instance[CurveIndex::L].IsIdentity() || !instance[CurveIndex::a].IsIdentity() || !instance[CurveIndex::b].IsIdentity() || !instance[CurveIndex::c].IsIdentity() )
++numberOfCurves;
if ( !instance[CurveIndex::H].IsIdentity() || !instance[CurveIndex::S].IsIdentity() )
++numberOfCurves;
}
if ( image.HasAlphaChannels() && !instance[CurveIndex::A].IsIdentity() )
++numberOfCurves;
if ( numberOfCurves == 0 )
{
Console().WriteLn( "<end><cbr><* Identity *>" );
return;
}
size_type N = image.NumberOfPixels();
int numberOfThreads = Thread::NumberOfThreads( N, 256 );
size_type pixelsPerThread = N/numberOfThreads;
image.Status().Initialize( "Curves transformation", numberOfCurves*N );
ThreadData data( image, numberOfCurves*N );
if ( useLUT )
data.lut.Generate( image, instance );
ReferenceArray<CurvesThread<P> > threads;
for ( int i = 0, j = 1; i < numberOfThreads; ++i, ++j )
threads.Add( new CurvesThread<P>( instance, data, image,
i*pixelsPerThread,
(j < numberOfThreads) ? j*pixelsPerThread : N ) );
AbstractImage::RunThreads( threads, data );
threads.Destroy();
image.Status() = data.status;
}
static void Apply( GenericImage<P>& image, const LocalHistogramEqualizationInstance& instance )
{
if ( image.IsColor() )
{
Image L;
image.GetLightness( L );
L.Status() = image.Status();
Apply( L, instance );
image.Status() = L.Status();
image.SetLightness( L );
return;
}
// create copy of the luminance to evaluate histogram from
GenericImage<P> imageCopy( image );
imageCopy.EnsureUnique(); // really not necessary, but we'll be safer if this is done
size_type N = image.NumberOfPixels();
int numberOfThreads = Thread::NumberOfThreads( image.Height(), 1 );
int rowsPerThread = image.Height()/numberOfThreads;
image.Status().Initialize( "CLAHE", N );
AbstractImage::ThreadData data( image, N );
// create processing threads
ReferenceArray<LocalHistogramEqualizationThread<P> > threads;
for ( int i = 0, j = 1; i < numberOfThreads; ++i, ++j )
threads.Add( new LocalHistogramEqualizationThread<P>( data,
instance,
image,
imageCopy,
i*rowsPerThread,
(j < numberOfThreads) ? j*rowsPerThread : image.Height() ) );
AbstractImage::RunThreads( threads, data );
threads.Destroy();
image.Status() = data.status;
}
template <class P, class S> static
void Apply( GenericImage<P>& image, S*,
const ICCProfileTransformation& T,
ICCProfileTransformation::transformation_handle transformation )
{
if ( image.IsEmptySelection() || T.Profiles().IsEmpty() )
return;
if ( image.ColorSpace() != ColorSpace::RGB && image.ColorSpace() != ColorSpace::Gray )
throw Error( String().Format( "Unsupported color space %X in ICC color transformation.", image.ColorSpace() ) );
image.EnsureUnique();
Rect r = image.SelectedRectangle();
int h = r.Height();
int numberOfThreads = T.IsParallelProcessingEnabled() ? Min( T.MaxProcessors(), pcl::Thread::NumberOfThreads( h, 1 ) ) : 1;
int rowsPerThread = h/numberOfThreads;
size_type N = image.NumberOfSelectedPixels();
if ( image.Status().IsInitializationEnabled() )
image.Status().Initialize( "In-place ICC color profile transformation", N );
ThreadData<P> data( image, T, transformation, N );
ReferenceArray<Thread<P,S> > threads;
for ( int i = 0, j = 1; i < numberOfThreads; ++i, ++j )
threads.Add( new Thread<P,S>( data,
i*rowsPerThread,
(j < numberOfThreads) ? j*rowsPerThread : h ) );
AbstractImage::RunThreads( threads, data );
threads.Destroy();
image.Status() = data.status;
}
template <class P> static
void Apply( GenericImage<P>& image, const MorphologicalTransformation& transformation )
{
if ( image.IsEmptySelection() )
return;
image.EnsureUnique();
int n = transformation.OverlappingDistance();
if ( n > image.Height() || n > image.Width() )
{
image.Zero();
return;
}
/*
* Dilation requires a reflected structure. We'll unreflect it once the
* transformation has finished.
*/
bool didReflect = false;
if ( transformation.Operator().IsDilation() != transformation.Structure().IsReflected() )
{
const_cast<StructuringElement&>( transformation.Structure() ).Reflect();
didReflect = true;
}
int numberOfRows = image.SelectedRectangle().Height();
int numberOfThreads = transformation.IsParallelProcessingEnabled() ?
Min( transformation.MaxProcessors(), pcl::Thread::NumberOfThreads( numberOfRows, n ) ) : 1;
int rowsPerThread = numberOfRows/numberOfThreads;
size_type N = image.NumberOfSelectedSamples();
if ( image.Status().IsInitializationEnabled() )
image.Status().Initialize( "Morphological transformation, " + transformation.Operator().Description(), N );
ThreadData<P> data( image, transformation, N );
ReferenceArray<Thread<P> > threads;
for ( int i = 0, j = 1, y0 = image.SelectedRectangle().y0; i < numberOfThreads; ++i, ++j )
threads.Add( new Thread<P>( data,
y0 + i*rowsPerThread,
y0 + ((j < numberOfThreads) ? j*rowsPerThread : numberOfRows),
i > 0,
j < numberOfThreads ) );
try
{
AbstractImage::RunThreads( threads, data );
if ( didReflect )
const_cast<StructuringElement&>( transformation.Structure() ).Reflect();
}
catch ( ... )
{
if ( didReflect )
const_cast<StructuringElement&>( transformation.Structure() ).Reflect();
throw;
}
image.SetStatusCallback( nullptr );
int c0 = image.SelectedChannel();
Point p0 = image.SelectedRectangle().LeftTop();
for ( int i = 0, j = 1; i < numberOfThreads; ++i, ++j )
{
if ( i > 0 )
image.Mov( threads[i].UpperOverlappingRegion(),
Point( p0.x, p0.y + i*rowsPerThread ), c0 );
if ( j < numberOfThreads )
image.Mov( threads[i].LowerOverlappingRegion(),
Point( p0.x, p0.y + j*rowsPerThread - threads[i].LowerOverlappingRegion().Height() ), c0 );
}
image.Status() = data.status;
threads.Destroy();
}
template <class P> static
void Apply( GenericImage<P>& image, const Translation& translation )
{
if ( translation.Delta() == 0.0 )
return;
int width = image.Width();
int height = image.Height();
if ( width == 0 || height == 0 )
return;
image.EnsureUnique();
typename P::sample* f = nullptr;
typename P::sample** f0 = nullptr;
int n = image.NumberOfChannels();
typename GenericImage<P>::color_space cs0 = image.ColorSpace();
StatusMonitor status = image.Status();
int numberOfThreads = translation.IsParallelProcessingEnabled() ?
Min( translation.MaxProcessors(), pcl::Thread::NumberOfThreads( height, 1 ) ) : 1;
int rowsPerThread = height/numberOfThreads;
try
{
size_type N = size_type( width )*size_type( height );
if ( status.IsInitializationEnabled() )
status.Initialize( String().Format( "Translate dx=%.3lf, dy=%.3lf, ",
translation.Delta().x, translation.Delta().y ) + translation.Interpolation().Description(),
size_type( n )*N );
f0 = image.ReleaseData();
for ( int c = 0; c < n; ++c )
{
ThreadData<P> data( translation.Delta(), width, height, status, N );
data.f = f = image.Allocator().AllocatePixels( size_type( width )*size_type( height ) );
data.fillValue = (c < translation.FillValues().Length()) ? P::ToSample( translation.FillValues()[c] ) : P::MinSampleValue();
ReferenceArray<Thread<P> > threads;
for ( int i = 0, j = 1; i < numberOfThreads; ++i, ++j )
threads.Add( new Thread<P>( data,
translation.Interpolation().NewInterpolator<P>( f0[c], width, height ),
i*rowsPerThread,
(j < numberOfThreads) ? j*rowsPerThread : height ) );
AbstractImage::RunThreads( threads, data );
threads.Destroy();
image.Allocator().Deallocate( f0[c] );
f0[c] = f;
f = nullptr;
status = data.status;
}
image.ImportData( f0, width, height, n, cs0 ).Status() = status;
}
catch ( ... )
{
if ( f != nullptr )
image.Allocator().Deallocate( f );
if ( f0 != nullptr )
{
for ( int c = 0; c < n; ++c )
if ( f0[c] != nullptr )
image.Allocator().Deallocate( f0[c] );
image.Allocator().Deallocate( f0 );
}
image.FreeData();
throw;
}
}
template <class P> static
void Apply( GenericImage<P>& image, const Resample& resample )
{
int width = image.Width();
int w0 = width;
int height = image.Height();
int h0 = height;
resample.GetNewSizes( width, height );
if ( width == w0 && height == h0 )
return;
if ( width <= 0 || height <= 0 )
{
image.FreeData();
return;
}
image.EnsureUnique();
typename P::sample* f = nullptr;
typename P::sample** f0 = nullptr;
int n = image.NumberOfChannels();
typename GenericImage<P>::color_space cs0 = image.ColorSpace();
double rx = double( w0 )/width;
double ry = double( h0 )/height;
StatusMonitor status = image.Status();
int numberOfThreads = resample.IsParallelProcessingEnabled() ?
Min( resample.MaxProcessors(), pcl::Thread::NumberOfThreads( height, 1 ) ) : 1;
int rowsPerThread = height/numberOfThreads;
try
{
size_type N = size_type( width )*size_type( height );
if ( status.IsInitializationEnabled() )
status.Initialize( String().Format( "Resampling to %dx%d px, ", width, height )
+ resample.Interpolation().Description(), size_type( n )*N );
f0 = image.ReleaseData();
for ( int c = 0; c < n; ++c )
{
ThreadData<P> data( rx, ry, width, status, N );
data.f = f = image.Allocator().AllocatePixels( width, height );
ReferenceArray<Thread<P> > threads;
for ( int i = 0, j = 1; i < numberOfThreads; ++i, ++j )
threads.Add( new Thread<P>( data, resample.Interpolation().NewInterpolator<P>( f0[c], w0, h0 ),
i*rowsPerThread,
(j < numberOfThreads) ? j*rowsPerThread : height ) );
AbstractImage::RunThreads( threads, data );
threads.Destroy();
image.Allocator().Deallocate( f0[c] );
f0[c] = f;
f = nullptr;
status = data.status;
}
image.ImportData( f0, width, height, n, cs0 ).Status() = status;
}
catch ( ... )
{
if ( f != nullptr )
image.Allocator().Deallocate( f );
if ( f0 != nullptr )
{
for ( int c = 0; c < n; ++c )
if ( f0[c] != nullptr )
image.Allocator().Deallocate( f0[c] );
image.Allocator().Deallocate( f0 );
}
image.FreeData();
throw;
}
}