Mercurial > hg > orthanc-stone
view OrthancStone/Sources/Toolbox/FiniteProjectiveCamera.cpp @ 2031:a56f7ed0cdf9 deep-learning
integration mainline->deep-learning
author | Sebastien Jodogne <s.jodogne@gmail.com> |
---|---|
date | Fri, 23 Dec 2022 17:51:07 +0100 |
parents | 7053b8a0aaec |
children | 07964689cb0b |
line wrap: on
line source
/** * Stone of Orthanc * Copyright (C) 2012-2016 Sebastien Jodogne, Medical Physics * Department, University Hospital of Liege, Belgium * Copyright (C) 2017-2022 Osimis S.A., Belgium * Copyright (C) 2021-2022 Sebastien Jodogne, ICTEAM UCLouvain, Belgium * * This program is free software: you can redistribute it and/or * modify it under the terms of the GNU Lesser General Public License * as published by the Free Software Foundation, either version 3 of * the License, or (at your option) any later version. * * This program is distributed in the hope that it will be useful, but * WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * Lesser General Public License for more details. * * You should have received a copy of the GNU Lesser General Public * License along with this program. If not, see * <http://www.gnu.org/licenses/>. **/ #include "FiniteProjectiveCamera.h" #include "GeometryToolbox.h" #include "SubpixelReader.h" #include <Logging.h> #include <OrthancException.h> #include <Images/Image.h> #include <Images/ImageProcessing.h> namespace OrthancStone { void FiniteProjectiveCamera::ComputeMInverse() { using namespace boost::numeric::ublas; // inv(M) = inv(K * R) = inv(R) * inv(K) = R' * inv(K). This // matrix is always invertible, by definition of finite // projective cameras (page 157). Matrix kinv; LinearAlgebra::InvertUpperTriangularMatrix(kinv, k_); minv_ = prod(trans(r_), kinv); } void FiniteProjectiveCamera::Setup(const Matrix& k, const Matrix& r, const Vector& c) { if (k.size1() != 3 || k.size2() != 3 || !LinearAlgebra::IsCloseToZero(k(1, 0)) || !LinearAlgebra::IsCloseToZero(k(2, 0)) || !LinearAlgebra::IsCloseToZero(k(2, 1))) { LOG(ERROR) << "Invalid intrinsic parameters"; throw Orthanc::OrthancException(Orthanc::ErrorCode_ParameterOutOfRange); } if (r.size1() != 3 || r.size2() != 3) { LOG(ERROR) << "Invalid size for a 3D rotation matrix"; throw Orthanc::OrthancException(Orthanc::ErrorCode_ParameterOutOfRange); } if (!LinearAlgebra::IsRotationMatrix(r, 100.0 * std::numeric_limits<float>::epsilon())) { LOG(ERROR) << "Invalid rotation matrix"; throw Orthanc::OrthancException(Orthanc::ErrorCode_ParameterOutOfRange); } if (c.size() != 3) { LOG(ERROR) << "Invalid camera center"; throw Orthanc::OrthancException(Orthanc::ErrorCode_ParameterOutOfRange); } k_ = k; r_ = r; c_ = c; ComputeMInverse(); Matrix tmp = LinearAlgebra::IdentityMatrix(3); tmp.resize(3, 4); tmp(0, 3) = -c[0]; tmp(1, 3) = -c[1]; tmp(2, 3) = -c[2]; p_ = LinearAlgebra::Product(k, r, tmp); assert(p_.size1() == 3 && p_.size2() == 4); } void FiniteProjectiveCamera::Setup(const Matrix& p) { if (p.size1() != 3 || p.size2() != 4) { LOG(ERROR) << "Invalid camera matrix"; throw Orthanc::OrthancException(Orthanc::ErrorCode_ParameterOutOfRange); } p_ = p; // M is the left 3x3 submatrix of "P" Matrix m = p; m.resize(3, 3); // p4 is the last column of "P" Vector p4(3); p4[0] = p(0, 3); p4[1] = p(1, 3); p4[2] = p(2, 3); // The RQ decomposition is explained on page 157 LinearAlgebra::RQDecomposition3x3(k_, r_, m); ComputeMInverse(); c_ = LinearAlgebra::Product(-minv_, p4); } FiniteProjectiveCamera::FiniteProjectiveCamera(const double k[9], const double r[9], const double c[3]) { Matrix kk, rr; Vector cc; LinearAlgebra::FillMatrix(kk, 3, 3, k); LinearAlgebra::FillMatrix(rr, 3, 3, r); LinearAlgebra::FillVector(cc, 3, c); Setup(kk, rr, cc); } FiniteProjectiveCamera::FiniteProjectiveCamera(const double p[12]) { Matrix pp; LinearAlgebra::FillMatrix(pp, 3, 4, p); Setup(pp); } Vector FiniteProjectiveCamera::GetRayDirection(double x, double y) const { // This derives from Equation (6.14) on page 162, taking "mu = // 1" and noticing that "-inv(M)*p4" corresponds to the camera // center in finite projective cameras // The (x,y) coordinates on the imaged plane, as an homogeneous vector Vector xx(3); xx[0] = x; xx[1] = y; xx[2] = 1.0; return boost::numeric::ublas::prod(minv_, xx); } static Vector SetupApply(const Vector& v, bool infinityAllowed) { if (v.size() == 3) { // Vector "v" in non-homogeneous coordinates, add the homogeneous component Vector vv; LinearAlgebra::AssignVector(vv, v[0], v[1], v[2], 1.0); return vv; } else if (v.size() == 4) { // Vector "v" is already in homogeneous coordinates if (!infinityAllowed && LinearAlgebra::IsCloseToZero(v[3])) { LOG(ERROR) << "Cannot apply a finite projective camera to a " << "point at infinity with this method"; throw Orthanc::OrthancException(Orthanc::ErrorCode_ParameterOutOfRange); } return v; } else { LOG(ERROR) << "The input vector must represent a point in 3D"; throw Orthanc::OrthancException(Orthanc::ErrorCode_ParameterOutOfRange); } } void FiniteProjectiveCamera::ApplyFinite(double& x, double& y, const Vector& v) const { Vector p = boost::numeric::ublas::prod(p_, SetupApply(v, false)); if (LinearAlgebra::IsCloseToZero(p[2])) { // Point at infinity: Should not happen with a finite input point throw Orthanc::OrthancException(Orthanc::ErrorCode_InternalError); } else { x = p[0] / p[2]; y = p[1] / p[2]; } } Vector FiniteProjectiveCamera::ApplyGeneral(const Vector& v) const { return boost::numeric::ublas::prod(p_, SetupApply(v, true)); } static Vector AddHomogeneousCoordinate(const Vector& p) { assert(p.size() == 3); return LinearAlgebra::CreateVector(p[0], p[1], p[2], 1); } FiniteProjectiveCamera::FiniteProjectiveCamera(const Vector& camera, const Vector& principalPoint, double angle, unsigned int imageWidth, unsigned int imageHeight, double pixelSpacingX, double pixelSpacingY) { if (camera.size() != 3 || principalPoint.size() != 3 || LinearAlgebra::IsCloseToZero(pixelSpacingX) || LinearAlgebra::IsCloseToZero(pixelSpacingY)) { throw Orthanc::OrthancException(Orthanc::ErrorCode_ParameterOutOfRange); } const double focal = boost::numeric::ublas::norm_2(camera - principalPoint); if (LinearAlgebra::IsCloseToZero(focal)) { LOG(ERROR) << "Camera lies on the image plane"; throw Orthanc::OrthancException(Orthanc::ErrorCode_ParameterOutOfRange); } Matrix a; GeometryToolbox::AlignVectorsWithRotation(a, camera - principalPoint, LinearAlgebra::CreateVector(0, 0, -1)); Matrix r = LinearAlgebra::Product(GeometryToolbox::CreateRotationMatrixAlongZ(angle), a); Matrix k = LinearAlgebra::ZeroMatrix(3, 3); k(0,0) = focal / pixelSpacingX; k(1,1) = focal / pixelSpacingY; k(0,2) = static_cast<double>(imageWidth) / 2.0; k(1,2) = static_cast<double>(imageHeight) / 2.0; k(2,2) = 1; Setup(k, r, camera); { // Sanity checks Vector v1 = LinearAlgebra::Product(p_, AddHomogeneousCoordinate(camera)); Vector v2 = LinearAlgebra::Product(p_, AddHomogeneousCoordinate(principalPoint)); if (!LinearAlgebra::IsCloseToZero(v1[2]) || // Camera is mapped to singularity LinearAlgebra::IsCloseToZero(v2[2])) { throw Orthanc::OrthancException(Orthanc::ErrorCode_InternalError); } // The principal point must be mapped to the center of the image v2 /= v2[2]; if (!LinearAlgebra::IsNear(v2[0], static_cast<double>(imageWidth) / 2.0) || !LinearAlgebra::IsNear(v2[1], static_cast<double>(imageHeight) / 2.0)) { throw Orthanc::OrthancException(Orthanc::ErrorCode_InternalError); } } } template <Orthanc::PixelFormat TargetFormat, Orthanc::PixelFormat SourceFormat, bool MIP> static void ApplyRaytracerInternal(Orthanc::ImageAccessor& target, const FiniteProjectiveCamera& camera, const ImageBuffer3D& source, const VolumeImageGeometry& geometry, VolumeProjection projection) { if (source.GetFormat() != SourceFormat || target.GetFormat() != TargetFormat || !std::numeric_limits<float>::is_iec559 || sizeof(float) != 4) { throw Orthanc::OrthancException(Orthanc::ErrorCode_InternalError); } LOG(WARNING) << "Input volume size: " << source.GetWidth() << "x" << source.GetHeight() << "x" << source.GetDepth(); LOG(WARNING) << "Input pixel format: " << Orthanc::EnumerationToString(source.GetFormat()); LOG(WARNING) << "Output image size: " << target.GetWidth() << "x" << target.GetHeight(); LOG(WARNING) << "Output pixel format: " << Orthanc::EnumerationToString(target.GetFormat()); const unsigned int slicesCount = geometry.GetProjectionDepth(projection); const Vector pixelSpacing = geometry.GetVoxelDimensions(projection); const unsigned int targetWidth = target.GetWidth(); const unsigned int targetHeight = target.GetHeight(); Orthanc::Image accumulator(Orthanc::PixelFormat_Float32, targetWidth, targetHeight, false); Orthanc::Image counter(Orthanc::PixelFormat_Grayscale16, targetWidth, targetHeight, false); Orthanc::ImageProcessing::Set(accumulator, 0); Orthanc::ImageProcessing::Set(counter, 0); typedef SubpixelReader<SourceFormat, ImageInterpolation_Nearest> SourceReader; for (unsigned int z = 0; z < slicesCount; z++) { LOG(INFO) << "Applying raytracer on slice: " << z << "/" << slicesCount; CoordinateSystem3D slice = geometry.GetProjectionSlice(projection, z); ImageBuffer3D::SliceReader sliceReader(source, projection, static_cast<unsigned int>(z)); SourceReader pixelReader(sliceReader.GetAccessor()); for (unsigned int y = 0; y < targetHeight; y++) { float *qacc = reinterpret_cast<float*>(accumulator.GetRow(y)); uint16_t *qcount = reinterpret_cast<uint16_t*>(counter.GetRow(y)); for (unsigned int x = 0; x < targetWidth; x++) { // Backproject the ray originating from the center of the target pixel Vector direction = camera.GetRayDirection(static_cast<double>(x + 0.5), static_cast<double>(y + 0.5)); // Compute the 3D intersection of the ray with the slice plane Vector p; if (slice.IntersectLine(p, camera.GetCenter(), direction)) { // Compute the 2D coordinates of the intersections, in slice coordinates double ix, iy; slice.ProjectPoint(ix, iy, p); ix /= pixelSpacing[0]; iy /= pixelSpacing[1]; // Read and accumulate the value of the pixel float pixel; if (pixelReader.GetFloatValue( pixel, static_cast<float>(ix), static_cast<float>(iy))) { if (MIP) { // MIP rendering if (*qcount == 0) { (*qacc) = pixel; (*qcount) = 1; } else if (pixel > *qacc) { (*qacc) = pixel; } } else { // Mean intensity (*qacc) += pixel; (*qcount) ++; } } } qacc++; qcount++; } } } typedef Orthanc::PixelTraits<TargetFormat> TargetTraits; // "Flatten" the accumulator image to create the target image for (unsigned int y = 0; y < targetHeight; y++) { const float *qacc = reinterpret_cast<const float*>(accumulator.GetConstRow(y)); const uint16_t *qcount = reinterpret_cast<const uint16_t*>(counter.GetConstRow(y)); typename TargetTraits::PixelType *p = reinterpret_cast<typename TargetTraits::PixelType*>(target.GetRow(y)); for (unsigned int x = 0; x < targetWidth; x++) { if (*qcount == 0) { TargetTraits::SetZero(*p); } else { TargetTraits::FloatToPixel(*p, *qacc / static_cast<float>(*qcount)); } p++; qacc++; qcount++; } } } Orthanc::ImageAccessor* FiniteProjectiveCamera::ApplyRaytracer(const ImageBuffer3D& source, const VolumeImageGeometry& geometry, Orthanc::PixelFormat targetFormat, unsigned int targetWidth, unsigned int targetHeight, bool mip) const { // TODO - We consider the axial projection of the volume, but we // should choose the projection that is the "most perpendicular" // to the line joining the camera center and the principal point const VolumeProjection projection = VolumeProjection_Axial; std::unique_ptr<Orthanc::ImageAccessor> target (new Orthanc::Image(targetFormat, targetWidth, targetHeight, false)); if (targetFormat == Orthanc::PixelFormat_Grayscale16 && source.GetFormat() == Orthanc::PixelFormat_Grayscale16 && mip) { ApplyRaytracerInternal<Orthanc::PixelFormat_Grayscale16, Orthanc::PixelFormat_Grayscale16, true> (*target, *this, source, geometry, projection); } else if (targetFormat == Orthanc::PixelFormat_Grayscale16 && source.GetFormat() == Orthanc::PixelFormat_Grayscale16 && !mip) { ApplyRaytracerInternal<Orthanc::PixelFormat_Grayscale16, Orthanc::PixelFormat_Grayscale16, false> (*target, *this, source, geometry, projection); } else { throw Orthanc::OrthancException(Orthanc::ErrorCode_NotImplemented); } return target.release(); } }