CPLErr VRTKernelFilteredSource::XMLInit( CPLXMLNode *psTree,
const char *pszVRTPath,
void* pUniqueHandle,
std::map<CPLString, GDALDataset*>& oMapSharedSources )
{
{
const CPLErr eErr = VRTFilteredSource::XMLInit( psTree, pszVRTPath,
pUniqueHandle,
oMapSharedSources );
if( eErr != CE_None )
return eErr;
}
const int nNewKernelSize = atoi(CPLGetXMLValue(psTree,"Kernel.Size","0"));
if( nNewKernelSize == 0 )
return CE_None;
char **papszCoefItems =
CSLTokenizeString( CPLGetXMLValue(psTree,"Kernel.Coefs","") );
const int nCoefs = CSLCount(papszCoefItems);
const bool bSquare = nCoefs == nNewKernelSize * nNewKernelSize;
const bool bSeparable = nCoefs == nNewKernelSize && nCoefs != 1;
if( !bSquare && !bSeparable )
{
CSLDestroy( papszCoefItems );
CPLError( CE_Failure, CPLE_AppDefined,
"Got wrong number of filter kernel coefficients (%s). "
"Expected %d or %d, got %d.",
CPLGetXMLValue(psTree,"Kernel.Coefs",""),
nNewKernelSize * nNewKernelSize, nNewKernelSize, nCoefs );
return CE_Failure;
}
double *padfNewCoefs = static_cast<double *>(
CPLMalloc( sizeof(double) * nCoefs ) );
for( int i = 0; i < nCoefs; i++ )
padfNewCoefs[i] = CPLAtof(papszCoefItems[i]);
const CPLErr eErr = SetKernel( nNewKernelSize, bSeparable, padfNewCoefs );
CPLFree( padfNewCoefs );
CSLDestroy( papszCoefItems );
SetNormalized( atoi(CPLGetXMLValue(psTree,"Kernel.normalized","0")) );
return eErr;
}
CPLErr VRTKernelFilteredSource::XMLInit( CPLXMLNode *psTree,
const char *pszVRTPath )
{
CPLErr eErr = VRTFilteredSource::XMLInit( psTree, pszVRTPath );
int nNewKernelSize, i, nCoefs;
double *padfNewCoefs;
if( eErr != CE_None )
return eErr;
nNewKernelSize = atoi(CPLGetXMLValue(psTree,"Kernel.Size","0"));
if( nNewKernelSize == 0 )
return CE_None;
char **papszCoefItems =
CSLTokenizeString( CPLGetXMLValue(psTree,"Kernel.Coefs","") );
nCoefs = CSLCount(papszCoefItems);
if( nCoefs != nNewKernelSize * nNewKernelSize )
{
CSLDestroy( papszCoefItems );
CPLError( CE_Failure, CPLE_AppDefined,
"Got wrong number of filter kernel coefficients (%s).\n"
"Expected %d, got %d.",
CPLGetXMLValue(psTree,"Kernel.Coefs",""),
nNewKernelSize * nNewKernelSize, nCoefs );
return CE_Failure;
}
padfNewCoefs = (double *) CPLMalloc(sizeof(double) * nCoefs);
for( i = 0; i < nCoefs; i++ )
padfNewCoefs[i] = CPLAtof(papszCoefItems[i]);
eErr = SetKernel( nNewKernelSize, padfNewCoefs );
CPLFree( padfNewCoefs );
CSLDestroy( papszCoefItems );
SetNormalized( atoi(CPLGetXMLValue(psTree,"Kernel.normalized","0")) );
return eErr;
}
int main(int argc, char **argv) {
args::ArgumentParser parser("Generates masks in stages.\n"
"Stage 1 - Otsu thresholding to generate binary mask\n"
"Stage 2 - RATs (optional)\n"
"Stage 3 - Hole filling (optional)\n"
"http://github.com/spinicist/QUIT");
args::Positional<std::string> input_path(parser, "INPUT_FILE", "Input file");
args::HelpFlag help(parser, "HELP", "Show this help menu", {'h', "help"});
args::Flag verbose(parser, "VERBOSE", "Print more information", {'v', "verbose"});
args::ValueFlag<std::string> outarg(
parser, "OUTPUT FILE", "Set output filename, default is input + _mask", {'o', "out"});
args::ValueFlag<int> volume(
parser, "VOLUME",
"Choose volume to mask in multi-volume file. Default 1, -1 selects last volume",
{'v', "volume"}, 0);
args::Flag is_complex(parser, "COMPLEX", "Input data is complex, take magnitude first",
{'x', "complex"});
args::ValueFlag<float> lower_threshold(
parser, "LOWER THRESHOLD",
"Specify lower intensity threshold for 1st stage, otherwise Otsu's method is used",
{'l', "lower"}, 0.);
args::ValueFlag<float> upper_threshold(
parser, "UPPER THRESHOLD",
"Specify upper intensity threshold for 1st stage, otherwise Otsu's method is used",
{'u', "upper"}, std::numeric_limits<float>::infinity());
args::ValueFlag<float> rats(
parser, "RATS", "Perform the RATS step, argument is size threshold for connected component",
{'r', "rats"}, 0.);
args::ValueFlag<int> fillh_radius(
parser, "FILL HOLES", "Fill holes in thresholded mask with radius N", {'F', "fillh"}, 0);
QI::ParseArgs(parser, argc, argv, verbose);
QI::SeriesF::Pointer vols = ITK_NULLPTR;
if (is_complex) {
vols = QI::ReadMagnitudeImage<QI::SeriesF>(QI::CheckPos(input_path), verbose);
} else {
vols = QI::ReadImage<QI::SeriesF>(QI::CheckPos(input_path), verbose);
}
std::string out_path;
if (outarg.Get() == "") {
out_path = QI::StripExt(input_path.Get()) + "_mask" + QI::OutExt();
} else {
out_path = outarg.Get();
}
// Extract one volume to process
auto region = vols->GetLargestPossibleRegion();
if (static_cast<size_t>(std::abs(volume.Get())) < region.GetSize()[3]) {
int volume_to_get = (region.GetSize()[3] + volume.Get()) % region.GetSize()[3];
QI::Log(verbose, "Masking volume {}...", volume_to_get);
region.GetModifiableIndex()[3] = volume_to_get;
} else {
QI::Fail("Specified mask volume was invalid {} ", volume.Get());
}
region.GetModifiableSize()[3] = 0;
auto vol = itk::ExtractImageFilter<QI::SeriesF, QI::VolumeF>::New();
vol->SetExtractionRegion(region);
vol->SetInput(vols);
vol->SetDirectionCollapseToSubmatrix();
vol->Update();
QI::VolumeF::Pointer intensity_image = vol->GetOutput();
intensity_image->DisconnectPipeline();
/*
* Stage 1 - Otsu or Threshold
*/
QI::VolumeI::Pointer mask_image = ITK_NULLPTR;
if (lower_threshold || upper_threshold) {
QI::Log(verbose, "Thresholding range: {}-{}", lower_threshold.Get(), upper_threshold.Get());
mask_image =
QI::ThresholdMask(intensity_image, lower_threshold.Get(), upper_threshold.Get());
} else {
QI::Log(verbose, "Generating Otsu mask");
mask_image = QI::OtsuMask(intensity_image);
}
/*
* Stage 2 - RATS
*/
if (rats) {
typedef itk::BinaryBallStructuringElement<int, 3> TBall;
typedef itk::BinaryErodeImageFilter<QI::VolumeI, QI::VolumeI, TBall> TErode;
typedef itk::BinaryDilateImageFilter<QI::VolumeI, QI::VolumeI, TBall> TDilate;
float mask_volume = std::numeric_limits<float>::infinity();
float voxel_volume = QI::VoxelVolume(mask_image);
QI::Log(verbose, "Voxel volume: {}", voxel_volume);
int radius = 0;
QI::VolumeI::Pointer mask_rats;
while (mask_volume > rats.Get()) {
radius++;
TBall ball;
TBall::SizeType radii;
radii.Fill(radius);
ball.SetRadius(radii);
ball.CreateStructuringElement();
TErode::Pointer erode = TErode::New();
erode->SetInput(mask_image);
erode->SetForegroundValue(1);
erode->SetBackgroundValue(0);
erode->SetKernel(ball);
erode->Update();
std::vector<float> kept_sizes = QI::FindLabels(erode->GetOutput(), 0, 1, mask_rats);
mask_volume = kept_sizes[0] * voxel_volume;
auto dilate = TDilate::New();
dilate->SetKernel(ball);
dilate->SetInput(mask_rats);
dilate->SetForegroundValue(1);
dilate->SetBackgroundValue(0);
dilate->Update();
mask_rats = dilate->GetOutput();
mask_rats->DisconnectPipeline();
QI::Log(verbose, "Ran RATS iteration, radius = {} volume = {}", radius, mask_volume);
}
mask_image = mask_rats;
}
/*
* Stage 3 - Hole Filling
*/
QI::VolumeI::Pointer finalMask = ITK_NULLPTR;
if (fillh_radius) {
QI::Log(verbose, "Filling holes");
auto fillHoles = itk::VotingBinaryIterativeHoleFillingImageFilter<QI::VolumeI>::New();
itk::VotingBinaryIterativeHoleFillingImageFilter<QI::VolumeI>::InputSizeType radius;
radius.Fill(fillh_radius.Get());
fillHoles->SetInput(mask_image);
fillHoles->SetRadius(radius);
fillHoles->SetMajorityThreshold(2); // Corresponds to (rad^3-1)/2 + 2 threshold
fillHoles->SetBackgroundValue(0);
fillHoles->SetForegroundValue(1);
fillHoles->SetMaximumNumberOfIterations(3);
fillHoles->Update();
mask_image = fillHoles->GetOutput();
mask_image->DisconnectPipeline();
}
QI::WriteImage(mask_image, out_path, verbose);
QI::Log(verbose, "Finished.");
return EXIT_SUCCESS;
}
