ImageEx EstimatorRender::render(const AContainer &group, AProcessWatcher *watcher) const {
auto planes_amount = group.image(0).size();
auto min_point = group.minPoint();
ProgressWrapper( watcher ).setTotal( planes_amount * iterations * group.count() );
auto est = AverageRender().render( group ); //Starting estimate
est.scaleFactor( upscale_factor );
auto beta = color::WHITE * this->beta / group.count();
for( unsigned c=0; c<planes_amount; ++c ){
for( int i=0; i<iterations; i++ ){
if( ProgressWrapper(watcher).shouldCancel() )
return {};
auto output_copy = est[c];
//Improve estimate
for( unsigned j=0; j<group.count(); j++, ProgressWrapper( watcher ).add() )
sign( output_copy, degrade( est[c], {group, j, c} ), group.image(j)[c], group.pos(j)-min_point
, beta, channelScale(group, j, c)*upscale_factor );
//Regularization
if( lambda > 0.0 )
regularize( est[c], output_copy, reg_size, alpha, beta, lambda );
else
est[c] = output_copy;
}
//DEBUG: See how close our model gets to the input data
for( unsigned j=0; j<group.count(); j++ ){
save( degrade( est[c], {group, j, c} ), "deg" + QString::number(c) + "-" + QString::number(j) );
save( group.image(j)[c], "low" + QString::number(c) + "-" + QString::number(j) );
}
}
return est;
}
static Point<bool> imagesMoves( const AContainer& container ){
if( container.count() == 0 )
return { false, false };
Point<bool> moves{ false, false };
auto base = container.pos( 0 );
for( unsigned i=1; i<container.count(); ++i ){
auto current = container.pos( i );
moves.x = moves.x || (base.x != current.x);
moves.y = moves.y || (base.y != current.y);
}
return moves;
}
void LinearAligner::align( AContainer& container, AProcessWatcher* watcher ) const {
LinearFunc hor, ver;//, both;
for( unsigned i=0; i<container.count(); i++ ){
hor.add( i, container.pos(i).x );
ver.add( i, container.pos(i).y );
//both.add( pos(i).x, pos(i).y );
}
for( unsigned i=0; i<container.count(); i++ ){
switch( method ){
case AlignMethod::BOTH: container.setPos( i, { hor(i), ver(i) } ); break;
case AlignMethod::VER: container.setPos( i, { 0, ver(i) } ); break;
case AlignMethod::HOR: container.setPos( i, { hor(i), 0 } ); break;
};
}
}
static void setLimit( QCustomPlot& plot, const AContainer& images, Point<bool> moves ){
auto min_point = images.minPoint();
auto max_point = images.maxPoint();
plot.xAxis->setRange( moves.x ? min_point.x : 0, moves.x ? max_point.x : images.count() );
plot.yAxis->setRange( moves.y ? min_point.y : 0, moves.y ? max_point.y : images.count() );
plot.yAxis->setLabel( moves.y ? QObject::tr("Y") : QObject::tr("Id") );
plot.xAxis->setLabel( moves.y ? QObject::tr("X") : QObject::tr("Id") );
}
ImageEx DiffRender::render( const AContainer& aligner, unsigned max_count, AProcessWatcher* watcher ) const{
if( max_count > aligner.count() )
max_count = aligner.count();
//Normal render
ImageEx avg = SimpleRender( SimpleRender::FOR_MERGING ).render( aligner, max_count );
//Find the smallest shared size
QSize size = aligner.size().size(); //No image is larger than the final result
for( unsigned i=0; i<max_count; i++ ){
size.setWidth( min( (unsigned)size.width(), aligner.image(i).get_width() ) );
size.setHeight( min( (unsigned)size.height(), aligner.image(i).get_height() ) );
}
//Create final output image based on the smallest size
ImageEx img( ImageEx::GRAY );
img.create( size.width(), size.height() );
Plane& output = img[0];
if( watcher )
watcher->setTotal( 1000 );
//Iterate over each pixel in the output image
for( unsigned iy=0; iy<output.get_height(); iy++ ){
if( watcher )
watcher->setCurrent( iy * 1000 / output.get_height() );
color_type* out = output.scan_line( iy );
for( unsigned ix=0; ix<output.get_width(); ix++ ){
//Set the pixel to the static difference of all the images until max_count
StaticDiff diff( aligner, avg, ix, iy );
for( unsigned j=0; j<max_count; j++ )
diff.add_image( j );
out[ix] = diff.result();
}
}
img.apply( &Plane::normalize );
return img;
}
virtual void align( AContainer& container, AProcessWatcher* ) const override{
//Validate input
auto max_frames = amount * repeats;
if( amount < 1 || repeats < 1 )
throw std::runtime_error( "Invalid arguments, no frames would be generated" );
if( max_frames < offset )
throw std::runtime_error( "Offset is bigger than frames per cycle" );
//Calculate offsets
for( unsigned i=0; i<container.count(); i++ )
container.setFrame( i, ((i + max_frames - offset)/repeats) % amount );
//Adding max_frames to be sure it is positive before MOD
}
ImageEx JpegConstrainerRender::render(const AContainer &group, AProcessWatcher *watcher) const {
//Get the one image we want to constrain
if( group.count() != 1 )
throw std::runtime_error( "Can only be used on one image" );
auto img = group.image(0);
//TODO: Check color types
Progress progress( "JpegConstrainerRender", img.size(), watcher );
progress.loopAll( [&]( int c ){
//Constrain it to the Coeff
unsigned change=0;
img[c] = degrader.planes[c].degradeFromJpegPlane( img[c], jpeg.planes[c], change );
} );
return img;
}
ImageEx AverageRender::render( const AContainer& aligner, AProcessWatcher* watcher ) const{
Timer t( "AverageRender::render()" );
//Abort if no images
if( aligner.count() == 0 ){
qWarning( "No images to render!" );
return ImageEx();
}
unsigned planes_amount = for_merging ? 1 : aligner.image(0).size();
ProgressWrapper( watcher ).setTotal( aligner.count() * planes_amount );
//Determine if we need to care about alpha per plane
bool use_plane_alpha = false;
for( unsigned i=0; i<aligner.count(); ++i )
if( aligner.alpha( i ) || aligner.imageMask( i ) >= 0 ){
use_plane_alpha = true;
break;
}
//Check for movement in both direction
auto movement = aligner.hasMovement();
if( movement.first && movement.second )
use_plane_alpha = true;
auto color_space = aligner.image(0).getColorSpace();
ImageEx img( for_merging ? color_space.changed( Transform::GRAY ) : color_space );
AlphaScales masks;
for( unsigned c=0; c<planes_amount; c++ ){
auto scale = upscale_chroma ? Point<double>(1,1)
: aligner.image( 0 )[c].getSize().to<double>() / aligner.image( 0 )[0].getSize().to<double>();
masks.addScale( aligner, scale );
}
auto min_point = aligner.minPoint();
auto full = aligner.size().size;
for( unsigned c=0; c<planes_amount; c++ ){
//Determine local size
auto scale = upscale_chroma ? Point<double>(1,1)
: aligner.image( 0 )[c].getSize().to<double>() / aligner.image( 0 )[0].getSize().to<double>();
//TODO: something is wrong with the rounding, chroma-channels are slightly off
SumPlane sum( (scale * full).ceil() );
sum.spacing = spacing;
sum.offset = offset;
for( unsigned j=0; j<aligner.count(); j++ ){
auto& image = aligner.image( j );
auto pos = (scale * (aligner.pos(j) - min_point)).round();
auto plane = getScaled( image[c], (scale * image[0].getSize()).round() );
const Plane& alpha_plane = masks.getAlpha( c, aligner.imageMask( j ), aligner.alpha( j ) );
if( use_plane_alpha && alpha_plane.valid() )
sum.addAlphaPlane( plane(), alpha_plane, pos );
else
sum.addPlane( plane(), pos );
ProgressWrapper( watcher ).add();
}
img.addPlane( sum.average() );
if( c == 0 && use_plane_alpha )
img.alpha_plane() = sum.alpha();
//TODO: what to do about the rest? We should try to fill in the gaps?
}
return img;
}