Mercurial > hg > orthanc
diff OrthancFramework/UnitTestsSources/FromDcmtkTests.cpp @ 4055:9214e3a7b0a2 framework
moving FromDcmtkTests.cpp from OrthancServer to OrthancFramework
author | Sebastien Jodogne <s.jodogne@gmail.com> |
---|---|
date | Thu, 11 Jun 2020 12:52:09 +0200 |
parents | OrthancServer/UnitTestsSources/FromDcmtkTests.cpp@05b8fd21089c |
children | 0953b3dc3261 |
line wrap: on
line diff
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/OrthancFramework/UnitTestsSources/FromDcmtkTests.cpp Thu Jun 11 12:52:09 2020 +0200 @@ -0,0 +1,1885 @@ +/** + * Orthanc - A Lightweight, RESTful DICOM Store + * Copyright (C) 2012-2016 Sebastien Jodogne, Medical Physics + * Department, University Hospital of Liege, Belgium + * Copyright (C) 2017-2020 Osimis S.A., Belgium + * + * This program is free software: you can redistribute it and/or + * modify it under the terms of the GNU General Public License as + * published by the Free Software Foundation, either version 3 of the + * License, or (at your option) any later version. + * + * In addition, as a special exception, the copyright holders of this + * program give permission to link the code of its release with the + * OpenSSL project's "OpenSSL" library (or with modified versions of it + * that use the same license as the "OpenSSL" library), and distribute + * the linked executables. You must obey the GNU General Public License + * in all respects for all of the code used other than "OpenSSL". If you + * modify file(s) with this exception, you may extend this exception to + * your version of the file(s), but you are not obligated to do so. If + * you do not wish to do so, delete this exception statement from your + * version. If you delete this exception statement from all source files + * in the program, then also delete it here. + * + * 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 + * General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program. If not, see <http://www.gnu.org/licenses/>. + **/ + + +#if ORTHANC_UNIT_TESTS_LINK_FRAMEWORK == 1 +# include <OrthancFramework.h> +#endif + +#include "gtest/gtest.h" + +#include "../Sources/Compatibility.h" +#include "../Sources/DicomNetworking/DicomFindAnswers.h" +#include "../Sources/DicomParsing/DicomModification.h" +#include "../Sources/DicomParsing/DicomWebJsonVisitor.h" +#include "../Sources/DicomParsing/FromDcmtkBridge.h" +#include "../Sources/DicomParsing/Internals/DicomImageDecoder.h" +#include "../Sources/DicomParsing/ToDcmtkBridge.h" +#include "../Sources/Endianness.h" +#include "../Sources/Images/Image.h" +#include "../Sources/Images/ImageBuffer.h" +#include "../Sources/Images/ImageProcessing.h" +#include "../Sources/Images/PngReader.h" +#include "../Sources/Images/PngWriter.h" +#include "../Sources/OrthancException.h" +#include "../Sources/SystemToolbox.h" +#include "../Resources/CodeGeneration/EncodingTests.h" + +#include <dcmtk/dcmdata/dcelem.h> +#include <dcmtk/dcmdata/dcdeftag.h> +#include <boost/algorithm/string/predicate.hpp> + +#if ORTHANC_ENABLE_PUGIXML == 1 +# include <pugixml.hpp> +#endif + +using namespace Orthanc; + +TEST(DicomFormat, Tag) +{ + ASSERT_EQ("PatientName", FromDcmtkBridge::GetTagName(DicomTag(0x0010, 0x0010), "")); + + DicomTag t = FromDcmtkBridge::ParseTag("SeriesDescription"); + ASSERT_EQ(0x0008, t.GetGroup()); + ASSERT_EQ(0x103E, t.GetElement()); + + t = FromDcmtkBridge::ParseTag("0020-e040"); + ASSERT_EQ(0x0020, t.GetGroup()); + ASSERT_EQ(0xe040, t.GetElement()); + + // Test ==() and !=() operators + ASSERT_TRUE(DICOM_TAG_PATIENT_ID == DicomTag(0x0010, 0x0020)); + ASSERT_FALSE(DICOM_TAG_PATIENT_ID != DicomTag(0x0010, 0x0020)); +} + + +TEST(DicomModification, Basic) +{ + DicomModification m; + m.SetupAnonymization(DicomVersion_2008); + //m.SetLevel(DicomRootLevel_Study); + //m.ReplacePlainString(DICOM_TAG_PATIENT_ID, "coucou"); + //m.ReplacePlainString(DICOM_TAG_PATIENT_NAME, "coucou"); + + ParsedDicomFile o(true); + o.SaveToFile("UnitTestsResults/anon.dcm"); + + for (int i = 0; i < 10; i++) + { + char b[1024]; + sprintf(b, "UnitTestsResults/anon%06d.dcm", i); + std::unique_ptr<ParsedDicomFile> f(o.Clone(false)); + if (i > 4) + o.ReplacePlainString(DICOM_TAG_SERIES_INSTANCE_UID, "coucou"); + m.Apply(*f); + f->SaveToFile(b); + } +} + + +TEST(DicomModification, Anonymization) +{ + ASSERT_EQ(DICOM_TAG_PATIENT_NAME, FromDcmtkBridge::ParseTag("PatientName")); + + const DicomTag privateTag(0x0045, 0x1010); + const DicomTag privateTag2(FromDcmtkBridge::ParseTag("0031-1020")); + ASSERT_TRUE(privateTag.IsPrivate()); + ASSERT_TRUE(privateTag2.IsPrivate()); + ASSERT_EQ(0x0031, privateTag2.GetGroup()); + ASSERT_EQ(0x1020, privateTag2.GetElement()); + + std::string s; + ParsedDicomFile o(true); + o.ReplacePlainString(DICOM_TAG_PATIENT_NAME, "coucou"); + ASSERT_FALSE(o.GetTagValue(s, privateTag)); + o.Insert(privateTag, "private tag", false, "OrthancCreator"); + ASSERT_TRUE(o.GetTagValue(s, privateTag)); + ASSERT_STREQ("private tag", s.c_str()); + + ASSERT_FALSE(o.GetTagValue(s, privateTag2)); + ASSERT_THROW(o.Replace(privateTag2, std::string("hello"), false, DicomReplaceMode_ThrowIfAbsent, "OrthancCreator"), OrthancException); + ASSERT_FALSE(o.GetTagValue(s, privateTag2)); + o.Replace(privateTag2, std::string("hello"), false, DicomReplaceMode_IgnoreIfAbsent, "OrthancCreator"); + ASSERT_FALSE(o.GetTagValue(s, privateTag2)); + o.Replace(privateTag2, std::string("hello"), false, DicomReplaceMode_InsertIfAbsent, "OrthancCreator"); + ASSERT_TRUE(o.GetTagValue(s, privateTag2)); + ASSERT_STREQ("hello", s.c_str()); + o.Replace(privateTag2, std::string("hello world"), false, DicomReplaceMode_InsertIfAbsent, "OrthancCreator"); + ASSERT_TRUE(o.GetTagValue(s, privateTag2)); + ASSERT_STREQ("hello world", s.c_str()); + + ASSERT_TRUE(o.GetTagValue(s, DICOM_TAG_PATIENT_NAME)); + ASSERT_FALSE(Toolbox::IsUuid(s)); + + DicomModification m; + m.SetupAnonymization(DicomVersion_2008); + m.Keep(privateTag); + + m.Apply(o); + + ASSERT_TRUE(o.GetTagValue(s, DICOM_TAG_PATIENT_NAME)); + ASSERT_TRUE(Toolbox::IsUuid(s)); + ASSERT_TRUE(o.GetTagValue(s, privateTag)); + ASSERT_STREQ("private tag", s.c_str()); + + m.SetupAnonymization(DicomVersion_2008); + m.Apply(o); + ASSERT_FALSE(o.GetTagValue(s, privateTag)); +} + + +#include <dcmtk/dcmdata/dcuid.h> + +TEST(DicomModification, Png) +{ + // Red dot in http://en.wikipedia.org/wiki/Data_URI_scheme (RGBA image) + std::string s = "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAUAAAAFCAYAAACNbyblAAAAHElEQVQI12P4//8/w38GIAXDIBKE0DHxgljNBAAO9TXL0Y4OHwAAAABJRU5ErkJggg=="; + + std::string m, cc; + ASSERT_TRUE(Toolbox::DecodeDataUriScheme(m, cc, s)); + + ASSERT_EQ("image/png", m); + + PngReader reader; + reader.ReadFromMemory(cc); + + ASSERT_EQ(5u, reader.GetHeight()); + ASSERT_EQ(5u, reader.GetWidth()); + ASSERT_EQ(PixelFormat_RGBA32, reader.GetFormat()); + + ParsedDicomFile o(true); + o.EmbedContent(s); + o.SaveToFile("UnitTestsResults/png1.dcm"); + + // Red dot, without alpha channel + s = "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAUAAAAFCAIAAAACDbGyAAAACXBIWXMAAAsTAAALEwEAmpwYAAAAB3RJTUUH3gUGDTcIn2+8BgAAACJJREFUCNdj/P//PwMjIwME/P/P+J8BBTAxEOL/R9Lx/z8AynoKAXOeiV8AAAAASUVORK5CYII="; + o.EmbedContent(s); + o.SaveToFile("UnitTestsResults/png2.dcm"); + + // Check box in Graylevel8 + s = "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAABAAAAAQCAAAAAA6mKC9AAAACXBIWXMAAAsTAAALEwEAmpwYAAAAB3RJTUUH3gUGDDcB53FulQAAAElJREFUGNNtj0sSAEEEQ1+U+185s1CtmRkblQ9CZldsKHJDk6DLGLJa6chjh0ooQmpjXMM86zPwydGEj6Ed/UGykkEM8X+p3u8/8LcOJIWLGeMAAAAASUVORK5CYII="; + o.EmbedContent(s); + //o.ReplacePlainString(DICOM_TAG_SOP_CLASS_UID, UID_DigitalXRayImageStorageForProcessing); + o.SaveToFile("UnitTestsResults/png3.dcm"); + + + { + // Gradient in Graylevel16 + + ImageBuffer img; + img.SetWidth(256); + img.SetHeight(256); + img.SetFormat(PixelFormat_Grayscale16); + + ImageAccessor accessor; + img.GetWriteableAccessor(accessor); + + uint16_t v = 0; + for (unsigned int y = 0; y < img.GetHeight(); y++) + { + uint16_t *p = reinterpret_cast<uint16_t*>(accessor.GetRow(y)); + for (unsigned int x = 0; x < img.GetWidth(); x++, p++, v++) + { + *p = v; + } + } + + o.EmbedImage(accessor); + o.SaveToFile("UnitTestsResults/png4.dcm"); + } +} + + +TEST(FromDcmtkBridge, Encodings1) +{ + for (unsigned int i = 0; i < testEncodingsCount; i++) + { + std::string source(testEncodingsEncoded[i]); + std::string expected(testEncodingsExpected[i]); + std::string s = Toolbox::ConvertToUtf8(source, testEncodings[i], false); + //std::cout << EnumerationToString(testEncodings[i]) << std::endl; + EXPECT_EQ(expected, s); + } +} + + +TEST(FromDcmtkBridge, Enumerations) +{ + // http://dicom.nema.org/medical/dicom/current/output/html/part03.html#sect_C.12.1.1.2 + Encoding e; + + ASSERT_FALSE(GetDicomEncoding(e, "")); + ASSERT_TRUE(GetDicomEncoding(e, "ISO_IR 6")); ASSERT_EQ(Encoding_Ascii, e); + + // http://dicom.nema.org/medical/dicom/current/output/html/part03.html#table_C.12-2 + ASSERT_TRUE(GetDicomEncoding(e, "ISO_IR 100")); ASSERT_EQ(Encoding_Latin1, e); + ASSERT_TRUE(GetDicomEncoding(e, "ISO_IR 101")); ASSERT_EQ(Encoding_Latin2, e); + ASSERT_TRUE(GetDicomEncoding(e, "ISO_IR 109")); ASSERT_EQ(Encoding_Latin3, e); + ASSERT_TRUE(GetDicomEncoding(e, "ISO_IR 110")); ASSERT_EQ(Encoding_Latin4, e); + ASSERT_TRUE(GetDicomEncoding(e, "ISO_IR 144")); ASSERT_EQ(Encoding_Cyrillic, e); + ASSERT_TRUE(GetDicomEncoding(e, "ISO_IR 127")); ASSERT_EQ(Encoding_Arabic, e); + ASSERT_TRUE(GetDicomEncoding(e, "ISO_IR 126")); ASSERT_EQ(Encoding_Greek, e); + ASSERT_TRUE(GetDicomEncoding(e, "ISO_IR 138")); ASSERT_EQ(Encoding_Hebrew, e); + ASSERT_TRUE(GetDicomEncoding(e, "ISO_IR 148")); ASSERT_EQ(Encoding_Latin5, e); + ASSERT_TRUE(GetDicomEncoding(e, "ISO_IR 13")); ASSERT_EQ(Encoding_Japanese, e); + ASSERT_TRUE(GetDicomEncoding(e, "ISO_IR 166")); ASSERT_EQ(Encoding_Thai, e); + + // http://dicom.nema.org/medical/dicom/current/output/html/part03.html#table_C.12-3 + ASSERT_TRUE(GetDicomEncoding(e, "ISO 2022 IR 6")); ASSERT_EQ(Encoding_Ascii, e); + ASSERT_TRUE(GetDicomEncoding(e, "ISO 2022 IR 100")); ASSERT_EQ(Encoding_Latin1, e); + ASSERT_TRUE(GetDicomEncoding(e, "ISO 2022 IR 101")); ASSERT_EQ(Encoding_Latin2, e); + ASSERT_TRUE(GetDicomEncoding(e, "ISO 2022 IR 109")); ASSERT_EQ(Encoding_Latin3, e); + ASSERT_TRUE(GetDicomEncoding(e, "ISO 2022 IR 110")); ASSERT_EQ(Encoding_Latin4, e); + ASSERT_TRUE(GetDicomEncoding(e, "ISO 2022 IR 144")); ASSERT_EQ(Encoding_Cyrillic, e); + ASSERT_TRUE(GetDicomEncoding(e, "ISO 2022 IR 127")); ASSERT_EQ(Encoding_Arabic, e); + ASSERT_TRUE(GetDicomEncoding(e, "ISO 2022 IR 126")); ASSERT_EQ(Encoding_Greek, e); + ASSERT_TRUE(GetDicomEncoding(e, "ISO 2022 IR 138")); ASSERT_EQ(Encoding_Hebrew, e); + ASSERT_TRUE(GetDicomEncoding(e, "ISO 2022 IR 148")); ASSERT_EQ(Encoding_Latin5, e); + ASSERT_TRUE(GetDicomEncoding(e, "ISO 2022 IR 13")); ASSERT_EQ(Encoding_Japanese, e); + ASSERT_TRUE(GetDicomEncoding(e, "ISO 2022 IR 166")); ASSERT_EQ(Encoding_Thai, e); + + // http://dicom.nema.org/medical/dicom/current/output/html/part03.html#table_C.12-4 + ASSERT_TRUE(GetDicomEncoding(e, "ISO 2022 IR 87")); ASSERT_EQ(Encoding_JapaneseKanji, e); + ASSERT_FALSE(GetDicomEncoding(e, "ISO 2022 IR 159")); //ASSERT_EQ(Encoding_JapaneseKanjiSupplementary, e); + ASSERT_TRUE(GetDicomEncoding(e, "ISO 2022 IR 149")); ASSERT_EQ(Encoding_Korean, e); + ASSERT_TRUE(GetDicomEncoding(e, "ISO 2022 IR 58")); ASSERT_EQ(Encoding_SimplifiedChinese, e); + + // http://dicom.nema.org/medical/dicom/current/output/html/part03.html#table_C.12-5 + ASSERT_TRUE(GetDicomEncoding(e, "ISO_IR 192")); ASSERT_EQ(Encoding_Utf8, e); + ASSERT_TRUE(GetDicomEncoding(e, "GB18030")); ASSERT_EQ(Encoding_Chinese, e); + ASSERT_TRUE(GetDicomEncoding(e, "GBK")); ASSERT_EQ(Encoding_Chinese, e); +} + + +TEST(FromDcmtkBridge, Encodings3) +{ + for (unsigned int i = 0; i < testEncodingsCount; i++) + { + //std::cout << EnumerationToString(testEncodings[i]) << std::endl; + std::string dicom; + + { + ParsedDicomFile f(true); + f.SetEncoding(testEncodings[i]); + + std::string s = Toolbox::ConvertToUtf8(testEncodingsEncoded[i], testEncodings[i], false); + f.Insert(DICOM_TAG_PATIENT_NAME, s, false, ""); + f.SaveToMemoryBuffer(dicom); + } + + if (testEncodings[i] != Encoding_Windows1251) + { + ParsedDicomFile g(dicom); + + if (testEncodings[i] != Encoding_Ascii) + { + bool hasCodeExtensions; + ASSERT_EQ(testEncodings[i], g.DetectEncoding(hasCodeExtensions)); + ASSERT_FALSE(hasCodeExtensions); + } + + std::string tag; + ASSERT_TRUE(g.GetTagValue(tag, DICOM_TAG_PATIENT_NAME)); + ASSERT_EQ(std::string(testEncodingsExpected[i]), tag); + } + } +} + + +TEST(FromDcmtkBridge, ValueRepresentation) +{ + ASSERT_EQ(ValueRepresentation_PersonName, + FromDcmtkBridge::LookupValueRepresentation(DICOM_TAG_PATIENT_NAME)); + ASSERT_EQ(ValueRepresentation_Date, + FromDcmtkBridge::LookupValueRepresentation(DicomTag(0x0008, 0x0020) /* StudyDate */)); + ASSERT_EQ(ValueRepresentation_Time, + FromDcmtkBridge::LookupValueRepresentation(DicomTag(0x0008, 0x0030) /* StudyTime */)); + ASSERT_EQ(ValueRepresentation_DateTime, + FromDcmtkBridge::LookupValueRepresentation(DicomTag(0x0008, 0x002a) /* AcquisitionDateTime */)); + ASSERT_EQ(ValueRepresentation_NotSupported, + FromDcmtkBridge::LookupValueRepresentation(DicomTag(0x0001, 0x0001) /* some private tag */)); +} + + + +static const DicomTag REFERENCED_STUDY_SEQUENCE(0x0008, 0x1110); +static const DicomTag REFERENCED_PATIENT_SEQUENCE(0x0008, 0x1120); + +static void CreateSampleJson(Json::Value& a) +{ + { + Json::Value b = Json::objectValue; + b["PatientName"] = "Hello"; + b["PatientID"] = "World"; + b["StudyDescription"] = "Toto"; + a.append(b); + } + + { + Json::Value b = Json::objectValue; + b["PatientName"] = "data:application/octet-stream;base64,SGVsbG8y"; // echo -n "Hello2" | base64 + b["PatientID"] = "World2"; + a.append(b); + } +} + + + +TEST(ParsedDicomFile, InsertReplaceStrings) +{ + ParsedDicomFile f(true); + + f.Insert(DICOM_TAG_PATIENT_NAME, "World", false, ""); + ASSERT_THROW(f.Insert(DICOM_TAG_PATIENT_ID, "Hello", false, ""), OrthancException); // Already existing tag + f.ReplacePlainString(DICOM_TAG_SOP_INSTANCE_UID, "Toto"); // (*) + f.ReplacePlainString(DICOM_TAG_SOP_CLASS_UID, "Tata"); // (**) + + std::string s; + ASSERT_TRUE(f.LookupTransferSyntax(s)); + // The default transfer syntax depends on the OS endianness + ASSERT_TRUE(s == GetTransferSyntaxUid(DicomTransferSyntax_LittleEndianExplicit) || + s == GetTransferSyntaxUid(DicomTransferSyntax_BigEndianExplicit)); + + ASSERT_THROW(f.Replace(DICOM_TAG_ACCESSION_NUMBER, std::string("Accession"), + false, DicomReplaceMode_ThrowIfAbsent, ""), OrthancException); + f.Replace(DICOM_TAG_ACCESSION_NUMBER, std::string("Accession"), false, DicomReplaceMode_IgnoreIfAbsent, ""); + ASSERT_FALSE(f.GetTagValue(s, DICOM_TAG_ACCESSION_NUMBER)); + f.Replace(DICOM_TAG_ACCESSION_NUMBER, std::string("Accession"), false, DicomReplaceMode_InsertIfAbsent, ""); + ASSERT_TRUE(f.GetTagValue(s, DICOM_TAG_ACCESSION_NUMBER)); + ASSERT_EQ(s, "Accession"); + f.Replace(DICOM_TAG_ACCESSION_NUMBER, std::string("Accession2"), false, DicomReplaceMode_IgnoreIfAbsent, ""); + ASSERT_TRUE(f.GetTagValue(s, DICOM_TAG_ACCESSION_NUMBER)); + ASSERT_EQ(s, "Accession2"); + f.Replace(DICOM_TAG_ACCESSION_NUMBER, std::string("Accession3"), false, DicomReplaceMode_ThrowIfAbsent, ""); + ASSERT_TRUE(f.GetTagValue(s, DICOM_TAG_ACCESSION_NUMBER)); + ASSERT_EQ(s, "Accession3"); + + ASSERT_TRUE(f.GetTagValue(s, DICOM_TAG_PATIENT_NAME)); + ASSERT_EQ(s, "World"); + ASSERT_TRUE(f.GetTagValue(s, DICOM_TAG_SOP_INSTANCE_UID)); + ASSERT_EQ(s, "Toto"); + ASSERT_TRUE(f.GetTagValue(s, DICOM_TAG_MEDIA_STORAGE_SOP_INSTANCE_UID)); // Implicitly modified by (*) + ASSERT_EQ(s, "Toto"); + ASSERT_TRUE(f.GetTagValue(s, DICOM_TAG_SOP_CLASS_UID)); + ASSERT_EQ(s, "Tata"); + ASSERT_TRUE(f.GetTagValue(s, DICOM_TAG_MEDIA_STORAGE_SOP_CLASS_UID)); // Implicitly modified by (**) + ASSERT_EQ(s, "Tata"); +} + + + + +TEST(ParsedDicomFile, InsertReplaceJson) +{ + ParsedDicomFile f(true); + + Json::Value a; + CreateSampleJson(a); + + ASSERT_FALSE(f.HasTag(REFERENCED_STUDY_SEQUENCE)); + f.Remove(REFERENCED_STUDY_SEQUENCE); // No effect + f.Insert(REFERENCED_STUDY_SEQUENCE, a, true, ""); + ASSERT_TRUE(f.HasTag(REFERENCED_STUDY_SEQUENCE)); + ASSERT_THROW(f.Insert(REFERENCED_STUDY_SEQUENCE, a, true, ""), OrthancException); + f.Remove(REFERENCED_STUDY_SEQUENCE); + ASSERT_FALSE(f.HasTag(REFERENCED_STUDY_SEQUENCE)); + f.Insert(REFERENCED_STUDY_SEQUENCE, a, true, ""); + ASSERT_TRUE(f.HasTag(REFERENCED_STUDY_SEQUENCE)); + + ASSERT_FALSE(f.HasTag(REFERENCED_PATIENT_SEQUENCE)); + ASSERT_THROW(f.Replace(REFERENCED_PATIENT_SEQUENCE, a, false, DicomReplaceMode_ThrowIfAbsent, ""), OrthancException); + ASSERT_FALSE(f.HasTag(REFERENCED_PATIENT_SEQUENCE)); + f.Replace(REFERENCED_PATIENT_SEQUENCE, a, false, DicomReplaceMode_IgnoreIfAbsent, ""); + ASSERT_FALSE(f.HasTag(REFERENCED_PATIENT_SEQUENCE)); + f.Replace(REFERENCED_PATIENT_SEQUENCE, a, false, DicomReplaceMode_InsertIfAbsent, ""); + ASSERT_TRUE(f.HasTag(REFERENCED_PATIENT_SEQUENCE)); + + { + Json::Value b; + f.DatasetToJson(b, DicomToJsonFormat_Full, DicomToJsonFlags_Default, 0); + + Json::Value c; + Toolbox::SimplifyDicomAsJson(c, b, DicomToJsonFormat_Human); + + ASSERT_EQ(0, c["ReferencedPatientSequence"].compare(a)); + ASSERT_NE(0, c["ReferencedStudySequence"].compare(a)); // Because Data URI Scheme decoding was enabled + } + + a = "data:application/octet-stream;base64,VGF0YQ=="; // echo -n "Tata" | base64 + f.Replace(DICOM_TAG_SOP_INSTANCE_UID, a, false, DicomReplaceMode_InsertIfAbsent, ""); // (*) + f.Replace(DICOM_TAG_SOP_CLASS_UID, a, true, DicomReplaceMode_InsertIfAbsent, ""); // (**) + + std::string s; + ASSERT_TRUE(f.GetTagValue(s, DICOM_TAG_SOP_INSTANCE_UID)); + ASSERT_EQ(s, a.asString()); + ASSERT_TRUE(f.GetTagValue(s, DICOM_TAG_MEDIA_STORAGE_SOP_INSTANCE_UID)); // Implicitly modified by (*) + ASSERT_EQ(s, a.asString()); + ASSERT_TRUE(f.GetTagValue(s, DICOM_TAG_SOP_CLASS_UID)); + ASSERT_EQ(s, "Tata"); + ASSERT_TRUE(f.GetTagValue(s, DICOM_TAG_MEDIA_STORAGE_SOP_CLASS_UID)); // Implicitly modified by (**) + ASSERT_EQ(s, "Tata"); +} + + +TEST(ParsedDicomFile, JsonEncoding) +{ + ParsedDicomFile f(true); + + for (unsigned int i = 0; i < testEncodingsCount; i++) + { + if (testEncodings[i] != Encoding_Windows1251) + { + //std::cout << EnumerationToString(testEncodings[i]) << std::endl; + f.SetEncoding(testEncodings[i]); + + if (testEncodings[i] != Encoding_Ascii) + { + bool hasCodeExtensions; + ASSERT_EQ(testEncodings[i], f.DetectEncoding(hasCodeExtensions)); + ASSERT_FALSE(hasCodeExtensions); + } + + Json::Value s = Toolbox::ConvertToUtf8(testEncodingsEncoded[i], testEncodings[i], false); + f.Replace(DICOM_TAG_PATIENT_NAME, s, false, DicomReplaceMode_InsertIfAbsent, ""); + + Json::Value v; + f.DatasetToJson(v, DicomToJsonFormat_Human, DicomToJsonFlags_Default, 0); + ASSERT_EQ(v["PatientName"].asString(), std::string(testEncodingsExpected[i])); + } + } +} + + +TEST(ParsedDicomFile, ToJsonFlags1) +{ + FromDcmtkBridge::RegisterDictionaryTag(DicomTag(0x7053, 0x1000), ValueRepresentation_OtherByte, "MyPrivateTag", 1, 1, "OrthancCreator"); + FromDcmtkBridge::RegisterDictionaryTag(DicomTag(0x7050, 0x1000), ValueRepresentation_PersonName, "Declared public tag", 1, 1, ""); + + ParsedDicomFile f(true); + f.Insert(DicomTag(0x7050, 0x1000), "Some public tag", false, ""); // Even group => public tag + f.Insert(DicomTag(0x7052, 0x1000), "Some unknown tag", false, ""); // Even group => public, unknown tag + f.Insert(DicomTag(0x7053, 0x1000), "Some private tag", false, "OrthancCreator"); // Odd group => private tag + + Json::Value v; + f.DatasetToJson(v, DicomToJsonFormat_Short, DicomToJsonFlags_None, 0); + ASSERT_EQ(Json::objectValue, v.type()); + ASSERT_EQ(6u, v.getMemberNames().size()); + ASSERT_FALSE(v.isMember("7052,1000")); + ASSERT_FALSE(v.isMember("7053,1000")); + ASSERT_TRUE(v.isMember("7050,1000")); + ASSERT_EQ(Json::stringValue, v["7050,1000"].type()); + ASSERT_EQ("Some public tag", v["7050,1000"].asString()); + + f.DatasetToJson(v, DicomToJsonFormat_Short, static_cast<DicomToJsonFlags>(DicomToJsonFlags_IncludePrivateTags | DicomToJsonFlags_IncludeBinary | DicomToJsonFlags_ConvertBinaryToNull), 0); + ASSERT_EQ(Json::objectValue, v.type()); + ASSERT_EQ(7u, v.getMemberNames().size()); + ASSERT_FALSE(v.isMember("7052,1000")); + ASSERT_TRUE(v.isMember("7050,1000")); + ASSERT_TRUE(v.isMember("7053,1000")); + ASSERT_EQ("Some public tag", v["7050,1000"].asString()); + ASSERT_EQ(Json::nullValue, v["7053,1000"].type()); + + f.DatasetToJson(v, DicomToJsonFormat_Short, static_cast<DicomToJsonFlags>(DicomToJsonFlags_IncludePrivateTags), 0); + ASSERT_EQ(Json::objectValue, v.type()); + ASSERT_EQ(6u, v.getMemberNames().size()); + ASSERT_FALSE(v.isMember("7052,1000")); + ASSERT_TRUE(v.isMember("7050,1000")); + ASSERT_FALSE(v.isMember("7053,1000")); + + f.DatasetToJson(v, DicomToJsonFormat_Short, static_cast<DicomToJsonFlags>(DicomToJsonFlags_IncludePrivateTags | DicomToJsonFlags_IncludeBinary), 0); + ASSERT_EQ(Json::objectValue, v.type()); + ASSERT_EQ(7u, v.getMemberNames().size()); + ASSERT_FALSE(v.isMember("7052,1000")); + ASSERT_TRUE(v.isMember("7050,1000")); + ASSERT_TRUE(v.isMember("7053,1000")); + ASSERT_EQ("Some public tag", v["7050,1000"].asString()); + std::string mime, content; + ASSERT_EQ(Json::stringValue, v["7053,1000"].type()); + ASSERT_TRUE(Toolbox::DecodeDataUriScheme(mime, content, v["7053,1000"].asString())); + ASSERT_EQ("application/octet-stream", mime); + ASSERT_EQ("Some private tag", content); + + f.DatasetToJson(v, DicomToJsonFormat_Short, static_cast<DicomToJsonFlags>(DicomToJsonFlags_IncludeUnknownTags | DicomToJsonFlags_IncludeBinary | DicomToJsonFlags_ConvertBinaryToNull), 0); + ASSERT_EQ(Json::objectValue, v.type()); + ASSERT_EQ(7u, v.getMemberNames().size()); + ASSERT_TRUE(v.isMember("7050,1000")); + ASSERT_TRUE(v.isMember("7052,1000")); + ASSERT_FALSE(v.isMember("7053,1000")); + ASSERT_EQ("Some public tag", v["7050,1000"].asString()); + ASSERT_EQ(Json::nullValue, v["7052,1000"].type()); + + f.DatasetToJson(v, DicomToJsonFormat_Short, static_cast<DicomToJsonFlags>(DicomToJsonFlags_IncludeUnknownTags | DicomToJsonFlags_IncludeBinary), 0); + ASSERT_EQ(Json::objectValue, v.type()); + ASSERT_EQ(7u, v.getMemberNames().size()); + ASSERT_TRUE(v.isMember("7050,1000")); + ASSERT_TRUE(v.isMember("7052,1000")); + ASSERT_FALSE(v.isMember("7053,1000")); + ASSERT_EQ("Some public tag", v["7050,1000"].asString()); + ASSERT_EQ(Json::stringValue, v["7052,1000"].type()); + ASSERT_TRUE(Toolbox::DecodeDataUriScheme(mime, content, v["7052,1000"].asString())); + ASSERT_EQ("application/octet-stream", mime); + ASSERT_EQ("Some unknown tag", content); + + f.DatasetToJson(v, DicomToJsonFormat_Short, static_cast<DicomToJsonFlags>(DicomToJsonFlags_IncludeUnknownTags | DicomToJsonFlags_IncludePrivateTags | DicomToJsonFlags_IncludeBinary | DicomToJsonFlags_ConvertBinaryToNull), 0); + ASSERT_EQ(Json::objectValue, v.type()); + ASSERT_EQ(8u, v.getMemberNames().size()); + ASSERT_TRUE(v.isMember("7050,1000")); + ASSERT_TRUE(v.isMember("7052,1000")); + ASSERT_TRUE(v.isMember("7053,1000")); + ASSERT_EQ("Some public tag", v["7050,1000"].asString()); + ASSERT_EQ(Json::nullValue, v["7052,1000"].type()); + ASSERT_EQ(Json::nullValue, v["7053,1000"].type()); +} + + +TEST(ParsedDicomFile, ToJsonFlags2) +{ + ParsedDicomFile f(true); + + { + // "ParsedDicomFile" uses Little Endian => 'B' (least significant + // byte) will be stored first in the memory buffer and in the + // file, then 'A'. Hence the expected "BA" value below. + Uint16 v[] = { 'A' * 256 + 'B', 0 }; + ASSERT_TRUE(f.GetDcmtkObject().getDataset()->putAndInsertUint16Array(DCM_PixelData, v, 2).good()); + } + + Json::Value v; + f.DatasetToJson(v, DicomToJsonFormat_Short, DicomToJsonFlags_None, 0); + ASSERT_EQ(Json::objectValue, v.type()); + ASSERT_EQ(5u, v.getMemberNames().size()); + ASSERT_FALSE(v.isMember("7fe0,0010")); + + f.DatasetToJson(v, DicomToJsonFormat_Short, static_cast<DicomToJsonFlags>(DicomToJsonFlags_IncludePixelData | DicomToJsonFlags_ConvertBinaryToNull), 0); + ASSERT_EQ(Json::objectValue, v.type()); + ASSERT_EQ(6u, v.getMemberNames().size()); + ASSERT_TRUE(v.isMember("7fe0,0010")); + ASSERT_EQ(Json::nullValue, v["7fe0,0010"].type()); + + f.DatasetToJson(v, DicomToJsonFormat_Short, static_cast<DicomToJsonFlags>(DicomToJsonFlags_IncludePixelData | DicomToJsonFlags_ConvertBinaryToAscii), 0); + ASSERT_EQ(Json::objectValue, v.type()); + ASSERT_EQ(6u, v.getMemberNames().size()); + ASSERT_TRUE(v.isMember("7fe0,0010")); + ASSERT_EQ(Json::stringValue, v["7fe0,0010"].type()); + ASSERT_EQ("BA", v["7fe0,0010"].asString().substr(0, 2)); + + f.DatasetToJson(v, DicomToJsonFormat_Short, DicomToJsonFlags_IncludePixelData, 0); + ASSERT_EQ(Json::objectValue, v.type()); + ASSERT_EQ(6u, v.getMemberNames().size()); + ASSERT_TRUE(v.isMember("7fe0,0010")); + ASSERT_EQ(Json::stringValue, v["7fe0,0010"].type()); + std::string mime, content; + ASSERT_TRUE(Toolbox::DecodeDataUriScheme(mime, content, v["7fe0,0010"].asString())); + ASSERT_EQ("application/octet-stream", mime); + ASSERT_EQ("BA", content.substr(0, 2)); +} + + +TEST(DicomFindAnswers, Basic) +{ + DicomFindAnswers a(false); + + { + DicomMap m; + m.SetValue(DICOM_TAG_PATIENT_ID, "hello", false); + a.Add(m); + } + + { + ParsedDicomFile d(true); + d.ReplacePlainString(DICOM_TAG_PATIENT_ID, "my"); + a.Add(d); + } + + { + DicomMap m; + m.SetValue(DICOM_TAG_PATIENT_ID, "world", false); + a.Add(m); + } + + Json::Value j; + a.ToJson(j, true); + ASSERT_EQ(3u, j.size()); + + //std::cout << j; +} + + +TEST(ParsedDicomFile, FromJson) +{ + FromDcmtkBridge::RegisterDictionaryTag(DicomTag(0x7057, 0x1000), ValueRepresentation_OtherByte, "MyPrivateTag2", 1, 1, "ORTHANC"); + FromDcmtkBridge::RegisterDictionaryTag(DicomTag(0x7059, 0x1000), ValueRepresentation_OtherByte, "MyPrivateTag3", 1, 1, ""); + FromDcmtkBridge::RegisterDictionaryTag(DicomTag(0x7050, 0x1000), ValueRepresentation_PersonName, "Declared public tag2", 1, 1, ""); + + Json::Value v; + const std::string sopClassUid = "1.2.840.10008.5.1.4.1.1.1"; // CR Image Storage: + + // Test the private creator + ASSERT_EQ(DcmTag_ERROR_TagName, FromDcmtkBridge::GetTagName(DicomTag(0x7057, 0x1000), "NOPE")); + ASSERT_EQ("MyPrivateTag2", FromDcmtkBridge::GetTagName(DicomTag(0x7057, 0x1000), "ORTHANC")); + + { + v["SOPClassUID"] = sopClassUid; + v["SpecificCharacterSet"] = "ISO_IR 148"; // This is latin-5 + v["PatientName"] = "Sébastien"; + v["7050-1000"] = "Some public tag"; // Even group => public tag + v["7052-1000"] = "Some unknown tag"; // Even group => public, unknown tag + v["7057-1000"] = "Some private tag"; // Odd group => private tag + v["7059-1000"] = "Some private tag2"; // Odd group => private tag, with an odd length to test padding + + std::string s; + Toolbox::EncodeDataUriScheme(s, "application/octet-stream", "Sebastien"); + v["StudyDescription"] = s; + + v["PixelData"] = "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAUAAAAFCAYAAACNbyblAAAAHElEQVQI12P4//8/w38GIAXDIBKE0DHxgljNBAAO9TXL0Y4OHwAAAABJRU5ErkJggg=="; // A red dot of 5x5 pixels + v["0040,0100"] = Json::arrayValue; // ScheduledProcedureStepSequence + + Json::Value vv; + vv["Modality"] = "MR"; + v["0040,0100"].append(vv); + + vv["Modality"] = "CT"; + v["0040,0100"].append(vv); + } + + const DicomToJsonFlags toJsonFlags = static_cast<DicomToJsonFlags>(DicomToJsonFlags_IncludeBinary | + DicomToJsonFlags_IncludePixelData | + DicomToJsonFlags_IncludePrivateTags | + DicomToJsonFlags_IncludeUnknownTags | + DicomToJsonFlags_ConvertBinaryToAscii); + + + { + std::unique_ptr<ParsedDicomFile> dicom + (ParsedDicomFile::CreateFromJson(v, static_cast<DicomFromJsonFlags>(DicomFromJsonFlags_GenerateIdentifiers), "")); + + Json::Value vv; + dicom->DatasetToJson(vv, DicomToJsonFormat_Human, toJsonFlags, 0); + + ASSERT_EQ(vv["SOPClassUID"].asString(), sopClassUid); + ASSERT_EQ(vv["MediaStorageSOPClassUID"].asString(), sopClassUid); + ASSERT_TRUE(vv.isMember("SOPInstanceUID")); + ASSERT_TRUE(vv.isMember("SeriesInstanceUID")); + ASSERT_TRUE(vv.isMember("StudyInstanceUID")); + ASSERT_TRUE(vv.isMember("PatientID")); + } + + + { + std::unique_ptr<ParsedDicomFile> dicom + (ParsedDicomFile::CreateFromJson(v, static_cast<DicomFromJsonFlags>(DicomFromJsonFlags_GenerateIdentifiers), "")); + + Json::Value vv; + dicom->DatasetToJson(vv, DicomToJsonFormat_Human, static_cast<DicomToJsonFlags>(DicomToJsonFlags_IncludePixelData), 0); + + std::string mime, content; + ASSERT_TRUE(Toolbox::DecodeDataUriScheme(mime, content, vv["PixelData"].asString())); + ASSERT_EQ("application/octet-stream", mime); + ASSERT_EQ(5u * 5u * 3u /* the red dot is 5x5 pixels in RGB24 */ + 1 /* for padding */, content.size()); + } + + + { + std::unique_ptr<ParsedDicomFile> dicom + (ParsedDicomFile::CreateFromJson(v, static_cast<DicomFromJsonFlags>(DicomFromJsonFlags_DecodeDataUriScheme), "")); + + Json::Value vv; + dicom->DatasetToJson(vv, DicomToJsonFormat_Short, toJsonFlags, 0); + + ASSERT_FALSE(vv.isMember("SOPInstanceUID")); + ASSERT_FALSE(vv.isMember("SeriesInstanceUID")); + ASSERT_FALSE(vv.isMember("StudyInstanceUID")); + ASSERT_FALSE(vv.isMember("PatientID")); + ASSERT_EQ(2u, vv["0040,0100"].size()); + ASSERT_EQ("MR", vv["0040,0100"][0]["0008,0060"].asString()); + ASSERT_EQ("CT", vv["0040,0100"][1]["0008,0060"].asString()); + ASSERT_EQ("Some public tag", vv["7050,1000"].asString()); + ASSERT_EQ("Some unknown tag", vv["7052,1000"].asString()); + ASSERT_EQ("Some private tag", vv["7057,1000"].asString()); + ASSERT_EQ("Some private tag2", vv["7059,1000"].asString()); + ASSERT_EQ("Sébastien", vv["0010,0010"].asString()); + ASSERT_EQ("Sebastien", vv["0008,1030"].asString()); + ASSERT_EQ("ISO_IR 148", vv["0008,0005"].asString()); + ASSERT_EQ("5", vv[DICOM_TAG_ROWS.Format()].asString()); + ASSERT_EQ("5", vv[DICOM_TAG_COLUMNS.Format()].asString()); + ASSERT_TRUE(vv[DICOM_TAG_PIXEL_DATA.Format()].asString().empty()); + } +} + + + +TEST(TestImages, PatternGrayscale8) +{ + static const char* PATH = "UnitTestsResults/PatternGrayscale8.dcm"; + + Orthanc::Image image(Orthanc::PixelFormat_Grayscale8, 256, 256, false); + + for (int y = 0; y < 256; y++) + { + uint8_t *p = reinterpret_cast<uint8_t*>(image.GetRow(y)); + for (int x = 0; x < 256; x++, p++) + { + *p = y; + } + } + + Orthanc::ImageAccessor r; + + image.GetRegion(r, 32, 32, 64, 192); + Orthanc::ImageProcessing::Set(r, 0); + + image.GetRegion(r, 160, 32, 64, 192); + Orthanc::ImageProcessing::Set(r, 255); + + { + ParsedDicomFile f(true); + f.ReplacePlainString(DICOM_TAG_SOP_CLASS_UID, "1.2.840.10008.5.1.4.1.1.7"); + f.ReplacePlainString(DICOM_TAG_STUDY_INSTANCE_UID, "1.2.276.0.7230010.3.1.2.2831176407.321.1458901422.884998"); + f.ReplacePlainString(DICOM_TAG_PATIENT_ID, "ORTHANC"); + f.ReplacePlainString(DICOM_TAG_PATIENT_NAME, "Orthanc"); + f.ReplacePlainString(DICOM_TAG_STUDY_DESCRIPTION, "Patterns"); + f.ReplacePlainString(DICOM_TAG_SERIES_DESCRIPTION, "Grayscale8"); + f.EmbedImage(image); + + f.SaveToFile(PATH); + } + + { + std::string s; + Orthanc::SystemToolbox::ReadFile(s, PATH); + Orthanc::ParsedDicomFile f(s); + + std::unique_ptr<Orthanc::ImageAccessor> decoded(Orthanc::DicomImageDecoder::Decode(f, 0)); + ASSERT_EQ(256u, decoded->GetWidth()); + ASSERT_EQ(256u, decoded->GetHeight()); + ASSERT_EQ(Orthanc::PixelFormat_Grayscale8, decoded->GetFormat()); + + for (int y = 0; y < 256; y++) + { + const void* a = image.GetConstRow(y); + const void* b = decoded->GetConstRow(y); + ASSERT_EQ(0, memcmp(a, b, 256)); + } + } +} + + +TEST(TestImages, PatternRGB) +{ + static const char* PATH = "UnitTestsResults/PatternRGB24.dcm"; + + Orthanc::Image image(Orthanc::PixelFormat_RGB24, 384, 256, false); + + for (int y = 0; y < 256; y++) + { + uint8_t *p = reinterpret_cast<uint8_t*>(image.GetRow(y)); + for (int x = 0; x < 128; x++, p += 3) + { + p[0] = y; + p[1] = 0; + p[2] = 0; + } + for (int x = 128; x < 128 * 2; x++, p += 3) + { + p[0] = 0; + p[1] = 255 - y; + p[2] = 0; + } + for (int x = 128 * 2; x < 128 * 3; x++, p += 3) + { + p[0] = 0; + p[1] = 0; + p[2] = y; + } + } + + { + ParsedDicomFile f(true); + f.ReplacePlainString(DICOM_TAG_SOP_CLASS_UID, "1.2.840.10008.5.1.4.1.1.7"); + f.ReplacePlainString(DICOM_TAG_STUDY_INSTANCE_UID, "1.2.276.0.7230010.3.1.2.2831176407.321.1458901422.884998"); + f.ReplacePlainString(DICOM_TAG_PATIENT_ID, "ORTHANC"); + f.ReplacePlainString(DICOM_TAG_PATIENT_NAME, "Orthanc"); + f.ReplacePlainString(DICOM_TAG_STUDY_DESCRIPTION, "Patterns"); + f.ReplacePlainString(DICOM_TAG_SERIES_DESCRIPTION, "RGB24"); + f.EmbedImage(image); + + f.SaveToFile(PATH); + } + + { + std::string s; + Orthanc::SystemToolbox::ReadFile(s, PATH); + Orthanc::ParsedDicomFile f(s); + + std::unique_ptr<Orthanc::ImageAccessor> decoded(Orthanc::DicomImageDecoder::Decode(f, 0)); + ASSERT_EQ(384u, decoded->GetWidth()); + ASSERT_EQ(256u, decoded->GetHeight()); + ASSERT_EQ(Orthanc::PixelFormat_RGB24, decoded->GetFormat()); + + for (int y = 0; y < 256; y++) + { + const void* a = image.GetConstRow(y); + const void* b = decoded->GetConstRow(y); + ASSERT_EQ(0, memcmp(a, b, 3 * 384)); + } + } +} + + +TEST(TestImages, PatternUint16) +{ + static const char* PATH = "UnitTestsResults/PatternGrayscale16.dcm"; + + Orthanc::Image image(Orthanc::PixelFormat_Grayscale16, 256, 256, false); + + uint16_t v = 0; + for (int y = 0; y < 256; y++) + { + uint16_t *p = reinterpret_cast<uint16_t*>(image.GetRow(y)); + for (int x = 0; x < 256; x++, v++, p++) + { + *p = htole16(v); // Orthanc uses Little-Endian transfer syntax to encode images + } + } + + Orthanc::ImageAccessor r; + + image.GetRegion(r, 32, 32, 64, 192); + Orthanc::ImageProcessing::Set(r, 0); + + image.GetRegion(r, 160, 32, 64, 192); + Orthanc::ImageProcessing::Set(r, 65535); + + { + ParsedDicomFile f(true); + f.ReplacePlainString(DICOM_TAG_SOP_CLASS_UID, "1.2.840.10008.5.1.4.1.1.7"); + f.ReplacePlainString(DICOM_TAG_STUDY_INSTANCE_UID, "1.2.276.0.7230010.3.1.2.2831176407.321.1458901422.884998"); + f.ReplacePlainString(DICOM_TAG_PATIENT_ID, "ORTHANC"); + f.ReplacePlainString(DICOM_TAG_PATIENT_NAME, "Orthanc"); + f.ReplacePlainString(DICOM_TAG_STUDY_DESCRIPTION, "Patterns"); + f.ReplacePlainString(DICOM_TAG_SERIES_DESCRIPTION, "Grayscale16"); + f.EmbedImage(image); + + f.SaveToFile(PATH); + } + + { + std::string s; + Orthanc::SystemToolbox::ReadFile(s, PATH); + Orthanc::ParsedDicomFile f(s); + + std::unique_ptr<Orthanc::ImageAccessor> decoded(Orthanc::DicomImageDecoder::Decode(f, 0)); + ASSERT_EQ(256u, decoded->GetWidth()); + ASSERT_EQ(256u, decoded->GetHeight()); + ASSERT_EQ(Orthanc::PixelFormat_Grayscale16, decoded->GetFormat()); + + for (int y = 0; y < 256; y++) + { + const void* a = image.GetConstRow(y); + const void* b = decoded->GetConstRow(y); + ASSERT_EQ(0, memcmp(a, b, 512)); + } + } +} + + +TEST(TestImages, PatternInt16) +{ + static const char* PATH = "UnitTestsResults/PatternSignedGrayscale16.dcm"; + + Orthanc::Image image(Orthanc::PixelFormat_SignedGrayscale16, 256, 256, false); + + int16_t v = -32768; + for (int y = 0; y < 256; y++) + { + int16_t *p = reinterpret_cast<int16_t*>(image.GetRow(y)); + for (int x = 0; x < 256; x++, v++, p++) + { + *p = htole16(v); // Orthanc uses Little-Endian transfer syntax to encode images + } + } + + Orthanc::ImageAccessor r; + image.GetRegion(r, 32, 32, 64, 192); + Orthanc::ImageProcessing::Set(r, -32768); + + image.GetRegion(r, 160, 32, 64, 192); + Orthanc::ImageProcessing::Set(r, 32767); + + { + ParsedDicomFile f(true); + f.ReplacePlainString(DICOM_TAG_SOP_CLASS_UID, "1.2.840.10008.5.1.4.1.1.7"); + f.ReplacePlainString(DICOM_TAG_STUDY_INSTANCE_UID, "1.2.276.0.7230010.3.1.2.2831176407.321.1458901422.884998"); + f.ReplacePlainString(DICOM_TAG_PATIENT_ID, "ORTHANC"); + f.ReplacePlainString(DICOM_TAG_PATIENT_NAME, "Orthanc"); + f.ReplacePlainString(DICOM_TAG_STUDY_DESCRIPTION, "Patterns"); + f.ReplacePlainString(DICOM_TAG_SERIES_DESCRIPTION, "SignedGrayscale16"); + f.EmbedImage(image); + + f.SaveToFile(PATH); + } + + { + std::string s; + Orthanc::SystemToolbox::ReadFile(s, PATH); + Orthanc::ParsedDicomFile f(s); + + std::unique_ptr<Orthanc::ImageAccessor> decoded(Orthanc::DicomImageDecoder::Decode(f, 0)); + ASSERT_EQ(256u, decoded->GetWidth()); + ASSERT_EQ(256u, decoded->GetHeight()); + ASSERT_EQ(Orthanc::PixelFormat_SignedGrayscale16, decoded->GetFormat()); + + for (int y = 0; y < 256; y++) + { + const void* a = image.GetConstRow(y); + const void* b = decoded->GetConstRow(y); + ASSERT_EQ(0, memcmp(a, b, 512)); + } + } +} + + + +static void CheckEncoding(const ParsedDicomFile& dicom, + Encoding expected) +{ + const char* value = NULL; + ASSERT_TRUE(dicom.GetDcmtkObject().getDataset()->findAndGetString(DCM_SpecificCharacterSet, value).good()); + + Encoding encoding; + ASSERT_TRUE(GetDicomEncoding(encoding, value)); + ASSERT_EQ(expected, encoding); +} + + +TEST(ParsedDicomFile, DicomMapEncodings1) +{ + SetDefaultDicomEncoding(Encoding_Ascii); + ASSERT_EQ(Encoding_Ascii, GetDefaultDicomEncoding()); + + { + DicomMap m; + ParsedDicomFile dicom(m, GetDefaultDicomEncoding(), false); + ASSERT_EQ(1u, dicom.GetDcmtkObject().getDataset()->card()); + CheckEncoding(dicom, Encoding_Ascii); + } + + { + DicomMap m; + ParsedDicomFile dicom(m, Encoding_Latin4, false); + ASSERT_EQ(1u, dicom.GetDcmtkObject().getDataset()->card()); + CheckEncoding(dicom, Encoding_Latin4); + } + + { + DicomMap m; + m.SetValue(DICOM_TAG_SPECIFIC_CHARACTER_SET, "ISO_IR 148", false); + ParsedDicomFile dicom(m, GetDefaultDicomEncoding(), false); + ASSERT_EQ(1u, dicom.GetDcmtkObject().getDataset()->card()); + CheckEncoding(dicom, Encoding_Latin5); + } + + { + DicomMap m; + m.SetValue(DICOM_TAG_SPECIFIC_CHARACTER_SET, "ISO_IR 148", false); + ParsedDicomFile dicom(m, Encoding_Latin1, false); + ASSERT_EQ(1u, dicom.GetDcmtkObject().getDataset()->card()); + CheckEncoding(dicom, Encoding_Latin5); + } +} + + +TEST(ParsedDicomFile, DicomMapEncodings2) +{ + const char* utf8 = NULL; + for (unsigned int i = 0; i < testEncodingsCount; i++) + { + if (testEncodings[i] == Encoding_Utf8) + { + utf8 = testEncodingsEncoded[i]; + break; + } + } + + ASSERT_TRUE(utf8 != NULL); + + for (unsigned int i = 0; i < testEncodingsCount; i++) + { + // 1251 codepage is not supported by the core DICOM standard, ignore it + if (testEncodings[i] != Encoding_Windows1251) + { + { + // Sanity check to test the proper behavior of "EncodingTests.py" + std::string encoded = Toolbox::ConvertFromUtf8(testEncodingsExpected[i], testEncodings[i]); + ASSERT_STREQ(testEncodingsEncoded[i], encoded.c_str()); + std::string decoded = Toolbox::ConvertToUtf8(encoded, testEncodings[i], false); + ASSERT_STREQ(testEncodingsExpected[i], decoded.c_str()); + + if (testEncodings[i] != Encoding_Chinese) + { + // A specific source string is used in "EncodingTests.py" to + // test against Chinese, it is normal that it does not correspond to UTF8 + + std::string encoded = Toolbox::ConvertToUtf8(Toolbox::ConvertFromUtf8(utf8, testEncodings[i]), testEncodings[i], false); + ASSERT_STREQ(testEncodingsExpected[i], encoded.c_str()); + } + } + + + Json::Value v; + + { + DicomMap m; + m.SetValue(DICOM_TAG_PATIENT_NAME, testEncodingsExpected[i], false); + + ParsedDicomFile dicom(m, testEncodings[i], false); + + const char* encoded = NULL; + ASSERT_TRUE(dicom.GetDcmtkObject().getDataset()->findAndGetString(DCM_PatientName, encoded).good()); + ASSERT_STREQ(testEncodingsEncoded[i], encoded); + + dicom.DatasetToJson(v, DicomToJsonFormat_Human, DicomToJsonFlags_Default, 0); + + Encoding encoding; + ASSERT_TRUE(GetDicomEncoding(encoding, v["SpecificCharacterSet"].asCString())); + ASSERT_EQ(encoding, testEncodings[i]); + ASSERT_STREQ(testEncodingsExpected[i], v["PatientName"].asCString()); + } + + + { + DicomMap m; + m.SetValue(DICOM_TAG_SPECIFIC_CHARACTER_SET, GetDicomSpecificCharacterSet(testEncodings[i]), false); + m.SetValue(DICOM_TAG_PATIENT_NAME, testEncodingsExpected[i], false); + + ParsedDicomFile dicom(m, testEncodings[i], false); + + Json::Value v2; + dicom.DatasetToJson(v2, DicomToJsonFormat_Human, DicomToJsonFlags_Default, 0); + + ASSERT_EQ(v2["PatientName"].asString(), v["PatientName"].asString()); + ASSERT_EQ(v2["SpecificCharacterSet"].asString(), v["SpecificCharacterSet"].asString()); + } + } + } +} + + +TEST(ParsedDicomFile, ChangeEncoding) +{ + for (unsigned int i = 0; i < testEncodingsCount; i++) + { + // 1251 codepage is not supported by the core DICOM standard, ignore it + if (testEncodings[i] != Encoding_Windows1251) + { + DicomMap m; + m.SetValue(DICOM_TAG_PATIENT_NAME, testEncodingsExpected[i], false); + + std::string tag; + + ParsedDicomFile dicom(m, Encoding_Utf8, false); + bool hasCodeExtensions; + ASSERT_EQ(Encoding_Utf8, dicom.DetectEncoding(hasCodeExtensions)); + ASSERT_FALSE(hasCodeExtensions); + ASSERT_TRUE(dicom.GetTagValue(tag, DICOM_TAG_PATIENT_NAME)); + ASSERT_EQ(tag, testEncodingsExpected[i]); + + { + Json::Value v; + dicom.DatasetToJson(v, DicomToJsonFormat_Human, DicomToJsonFlags_Default, 0); + ASSERT_STREQ(v["SpecificCharacterSet"].asCString(), "ISO_IR 192"); + ASSERT_STREQ(v["PatientName"].asCString(), testEncodingsExpected[i]); + } + + dicom.ChangeEncoding(testEncodings[i]); + + ASSERT_EQ(testEncodings[i], dicom.DetectEncoding(hasCodeExtensions)); + ASSERT_FALSE(hasCodeExtensions); + + const char* c = NULL; + ASSERT_TRUE(dicom.GetDcmtkObject().getDataset()->findAndGetString(DCM_PatientName, c).good()); + EXPECT_STREQ(c, testEncodingsEncoded[i]); + + ASSERT_TRUE(dicom.GetTagValue(tag, DICOM_TAG_PATIENT_NAME)); // Decodes to UTF-8 + EXPECT_EQ(tag, testEncodingsExpected[i]); + + { + Json::Value v; + dicom.DatasetToJson(v, DicomToJsonFormat_Human, DicomToJsonFlags_Default, 0); + ASSERT_STREQ(v["SpecificCharacterSet"].asCString(), GetDicomSpecificCharacterSet(testEncodings[i])); + ASSERT_STREQ(v["PatientName"].asCString(), testEncodingsExpected[i]); + } + } + } +} + + +TEST(Toolbox, CaseWithAccents) +{ + ASSERT_EQ(toUpperResult, Toolbox::ToUpperCaseWithAccents(toUpperSource)); +} + + + +TEST(ParsedDicomFile, InvalidCharacterSets) +{ + { + // No encoding provided, fallback to default encoding + DicomMap m; + m.SetValue(DICOM_TAG_PATIENT_NAME, "HELLO", false); + + ParsedDicomFile d(m, Encoding_Latin3 /* default encoding */, false); + + bool hasCodeExtensions; + ASSERT_EQ(Encoding_Latin3, d.DetectEncoding(hasCodeExtensions)); + ASSERT_FALSE(hasCodeExtensions); + } + + { + // Valid encoding, "ISO_IR 13" is Japanese + DicomMap m; + m.SetValue(DICOM_TAG_SPECIFIC_CHARACTER_SET, "ISO_IR 13", false); + m.SetValue(DICOM_TAG_PATIENT_NAME, "HELLO", false); + + ParsedDicomFile d(m, Encoding_Latin3 /* default encoding */, false); + + bool hasCodeExtensions; + ASSERT_EQ(Encoding_Japanese, d.DetectEncoding(hasCodeExtensions)); + ASSERT_FALSE(hasCodeExtensions); + } + + { + // Invalid value for an encoding ("nope" is not in the DICOM standard) + DicomMap m; + m.SetValue(DICOM_TAG_SPECIFIC_CHARACTER_SET, "nope", false); + m.SetValue(DICOM_TAG_PATIENT_NAME, "HELLO", false); + + ASSERT_THROW(ParsedDicomFile d(m, Encoding_Latin3, false), + OrthancException); + } + + { + // Invalid encoding, as provided as a binary string + DicomMap m; + m.SetValue(DICOM_TAG_SPECIFIC_CHARACTER_SET, "ISO_IR 13", true); + m.SetValue(DICOM_TAG_PATIENT_NAME, "HELLO", false); + + ASSERT_THROW(ParsedDicomFile d(m, Encoding_Latin3, false), + OrthancException); + } + + { + // Encoding provided as an empty string, fallback to default encoding + // In Orthanc <= 1.3.1, this test was throwing an exception + DicomMap m; + m.SetValue(DICOM_TAG_SPECIFIC_CHARACTER_SET, "", false); + m.SetValue(DICOM_TAG_PATIENT_NAME, "HELLO", false); + + ParsedDicomFile d(m, Encoding_Latin3 /* default encoding */, false); + + bool hasCodeExtensions; + ASSERT_EQ(Encoding_Latin3, d.DetectEncoding(hasCodeExtensions)); + ASSERT_FALSE(hasCodeExtensions); + } +} + + + +TEST(Toolbox, RemoveIso2022EscapeSequences) +{ + // +----------------------------------+ + // | one-byte control messages | + // +----------------------------------+ + + static const uint8_t iso2022_cstr_oneByteControl[] = { + 0x0f, 0x41, + 0x0e, 0x42, + 0x8e, 0x1b, 0x4e, 0x43, + 0x8f, 0x1b, 0x4f, 0x44, + 0x8e, 0x1b, 0x4a, 0x45, + 0x8f, 0x1b, 0x4a, 0x46, + 0x50, 0x51, 0x52, 0x00 + }; + + static const uint8_t iso2022_cstr_oneByteControl_ref[] = { + 0x41, + 0x42, + 0x43, + 0x44, + 0x8e, 0x1b, 0x4a, 0x45, + 0x8f, 0x1b, 0x4a, 0x46, + 0x50, 0x51, 0x52, 0x00 + }; + + // +----------------------------------+ + // | two-byte control messages | + // +----------------------------------+ + + static const uint8_t iso2022_cstr_twoByteControl[] = { + 0x1b, 0x6e, 0x41, + 0x1b, 0x6f, 0x42, + 0x1b, 0x4e, 0x43, + 0x1b, 0x4f, 0x44, + 0x1b, 0x7e, 0x45, + 0x1b, 0x7d, 0x46, + 0x1b, 0x7c, 0x47, 0x00 + }; + + static const uint8_t iso2022_cstr_twoByteControl_ref[] = { + 0x41, + 0x42, + 0x43, + 0x44, + 0x45, + 0x46, + 0x47, 0x00 + }; + + // +----------------------------------+ + // | various-length escape sequences | + // +----------------------------------+ + + static const uint8_t iso2022_cstr_escapeSequence[] = { + 0x1b, 0x40, 0x41, // 1b and 40 should not be removed (invalid esc seq) + 0x1b, 0x50, 0x42, // ditto + 0x1b, 0x7f, 0x43, // ditto + 0x1b, 0x21, 0x4a, 0x44, // this will match + 0x1b, 0x20, 0x21, 0x2f, 0x40, 0x45, // this will match + 0x1b, 0x20, 0x21, 0x2f, 0x2f, 0x40, 0x46, // this will match too + 0x1b, 0x20, 0x21, 0x2f, 0x1f, 0x47, 0x48, 0x00 // this will NOT match! + }; + + static const uint8_t iso2022_cstr_escapeSequence_ref[] = { + 0x1b, 0x40, 0x41, // 1b and 40 should not be removed (invalid esc seq) + 0x1b, 0x50, 0x42, // ditto + 0x1b, 0x7f, 0x43, // ditto + 0x44, // this will match + 0x45, // this will match + 0x46, // this will match too + 0x1b, 0x20, 0x21, 0x2f, 0x1f, 0x47, 0x48, 0x00 // this will NOT match! + }; + + + // +----------------------------------+ + // | a real-world japanese sample | + // +----------------------------------+ + + static const uint8_t iso2022_cstr_real_ir13[] = { + 0xd4, 0xcf, 0xc0, 0xde, 0x5e, 0xc0, 0xdb, 0xb3, + 0x3d, 0x1b, 0x24, 0x42, 0x3b, 0x33, 0x45, 0x44, + 0x1b, 0x28, 0x4a, 0x5e, 0x1b, 0x24, 0x42, 0x42, + 0x40, 0x4f, 0x3a, 0x1b, 0x28, 0x4a, 0x3d, 0x1b, + 0x24, 0x42, 0x24, 0x64, 0x24, 0x5e, 0x24, 0x40, + 0x1b, 0x28, 0x4a, 0x5e, 0x1b, 0x24, 0x42, 0x24, + 0x3f, 0x24, 0x6d, 0x24, 0x26, 0x1b, 0x28, 0x4a, 0x00 + }; + + static const uint8_t iso2022_cstr_real_ir13_ref[] = { + 0xd4, 0xcf, 0xc0, 0xde, 0x5e, 0xc0, 0xdb, 0xb3, + 0x3d, + 0x3b, 0x33, 0x45, 0x44, + 0x5e, + 0x42, + 0x40, 0x4f, 0x3a, + 0x3d, + 0x24, 0x64, 0x24, 0x5e, 0x24, 0x40, + 0x5e, + 0x24, + 0x3f, 0x24, 0x6d, 0x24, 0x26, 0x00 + }; + + + + // +----------------------------------+ + // | the actual test | + // +----------------------------------+ + + std::string iso2022_str_oneByteControl( + reinterpret_cast<const char*>(iso2022_cstr_oneByteControl)); + std::string iso2022_str_oneByteControl_ref( + reinterpret_cast<const char*>(iso2022_cstr_oneByteControl_ref)); + std::string iso2022_str_twoByteControl( + reinterpret_cast<const char*>(iso2022_cstr_twoByteControl)); + std::string iso2022_str_twoByteControl_ref( + reinterpret_cast<const char*>(iso2022_cstr_twoByteControl_ref)); + std::string iso2022_str_escapeSequence( + reinterpret_cast<const char*>(iso2022_cstr_escapeSequence)); + std::string iso2022_str_escapeSequence_ref( + reinterpret_cast<const char*>(iso2022_cstr_escapeSequence_ref)); + std::string iso2022_str_real_ir13( + reinterpret_cast<const char*>(iso2022_cstr_real_ir13)); + std::string iso2022_str_real_ir13_ref( + reinterpret_cast<const char*>(iso2022_cstr_real_ir13_ref)); + + std::string dest; + + Toolbox::RemoveIso2022EscapeSequences(dest, iso2022_str_oneByteControl); + ASSERT_EQ(dest, iso2022_str_oneByteControl_ref); + + Toolbox::RemoveIso2022EscapeSequences(dest, iso2022_str_twoByteControl); + ASSERT_EQ(dest, iso2022_str_twoByteControl_ref); + + Toolbox::RemoveIso2022EscapeSequences(dest, iso2022_str_escapeSequence); + ASSERT_EQ(dest, iso2022_str_escapeSequence_ref); + + Toolbox::RemoveIso2022EscapeSequences(dest, iso2022_str_real_ir13); + ASSERT_EQ(dest, iso2022_str_real_ir13_ref); +} + + + +static std::string DecodeFromSpecification(const std::string& s) +{ + std::vector<std::string> tokens; + Toolbox::TokenizeString(tokens, s, ' '); + + std::string result; + result.resize(tokens.size()); + + for (size_t i = 0; i < tokens.size(); i++) + { + std::vector<std::string> components; + Toolbox::TokenizeString(components, tokens[i], '/'); + + if (components.size() != 2) + { + throw; + } + + int a = boost::lexical_cast<int>(components[0]); + int b = boost::lexical_cast<int>(components[1]); + if (a < 0 || a > 15 || + b < 0 || b > 15 || + (a == 0 && b == 0)) + { + throw; + } + + result[i] = static_cast<uint8_t>(a * 16 + b); + } + + return result; +} + + + +// Compatibility wrapper +static pugi::xpath_node SelectNode(const pugi::xml_document& doc, + const char* xpath) +{ +#if PUGIXML_VERSION <= 140 + return doc.select_single_node(xpath); // Deprecated in pugixml 1.5 +#else + return doc.select_node(xpath); +#endif +} + + +TEST(Toolbox, EncodingsKorean) +{ + // http://dicom.nema.org/MEDICAL/dicom/current/output/chtml/part05/sect_I.2.html + + std::string korean = DecodeFromSpecification( + "04/08 06/15 06/14 06/07 05/14 04/07 06/09 06/12 06/04 06/15 06/14 06/07 03/13 " + "01/11 02/04 02/09 04/03 15/11 15/03 05/14 01/11 02/04 02/09 04/03 13/01 12/14 " + "13/04 13/07 03/13 01/11 02/04 02/09 04/03 12/08 10/11 05/14 01/11 02/04 02/09 " + "04/03 11/01 14/06 11/05 11/15"); + + // This array can be re-generated using command-line: + // echo -n "Hong^Gildong=..." | hexdump -v -e '14/1 "0x%02x, "' -e '"\n"' + static const uint8_t utf8raw[] = { + 0x48, 0x6f, 0x6e, 0x67, 0x5e, 0x47, 0x69, 0x6c, 0x64, 0x6f, 0x6e, 0x67, 0x3d, 0xe6, + 0xb4, 0xaa, 0x5e, 0xe5, 0x90, 0x89, 0xe6, 0xb4, 0x9e, 0x3d, 0xed, 0x99, 0x8d, 0x5e, + 0xea, 0xb8, 0xb8, 0xeb, 0x8f, 0x99 + }; + + std::string utf8(reinterpret_cast<const char*>(utf8raw), sizeof(utf8raw)); + + ParsedDicomFile dicom(false); + dicom.ReplacePlainString(DICOM_TAG_SPECIFIC_CHARACTER_SET, "\\ISO 2022 IR 149"); + ASSERT_TRUE(dicom.GetDcmtkObject().getDataset()->putAndInsertString + (DCM_PatientName, korean.c_str(), OFBool(true)).good()); + + bool hasCodeExtensions; + Encoding encoding = dicom.DetectEncoding(hasCodeExtensions); + ASSERT_EQ(Encoding_Korean, encoding); + ASSERT_TRUE(hasCodeExtensions); + + std::string value; + ASSERT_TRUE(dicom.GetTagValue(value, DICOM_TAG_PATIENT_NAME)); + ASSERT_EQ(utf8, value); + + DicomWebJsonVisitor visitor; + dicom.Apply(visitor); + ASSERT_EQ(utf8.substr(0, 12), visitor.GetResult()["00100010"]["Value"][0]["Alphabetic"].asString()); + ASSERT_EQ(utf8.substr(13, 10), visitor.GetResult()["00100010"]["Value"][0]["Ideographic"].asString()); + ASSERT_EQ(utf8.substr(24), visitor.GetResult()["00100010"]["Value"][0]["Phonetic"].asString()); + +#if ORTHANC_ENABLE_PUGIXML == 1 + // http://dicom.nema.org/medical/dicom/current/output/chtml/part18/sect_F.3.html#table_F.3.1-1 + std::string xml; + visitor.FormatXml(xml); + + pugi::xml_document doc; + doc.load_buffer(xml.c_str(), xml.size()); + + pugi::xpath_node node = SelectNode(doc, "//NativeDicomModel/DicomAttribute[@tag=\"00080005\"]/Value"); + ASSERT_STREQ("ISO 2022 IR 149", node.node().text().as_string()); + + node = SelectNode(doc, "//NativeDicomModel/DicomAttribute[@tag=\"00080005\"]"); + ASSERT_STREQ("CS", node.node().attribute("vr").value()); + + node = SelectNode(doc, "//NativeDicomModel/DicomAttribute[@tag=\"00100010\"]"); + ASSERT_STREQ("PN", node.node().attribute("vr").value()); + + node = SelectNode(doc, "//NativeDicomModel/DicomAttribute[@tag=\"00100010\"]/PersonName/Alphabetic/FamilyName"); + ASSERT_STREQ("Hong", node.node().text().as_string()); + + node = SelectNode(doc, "//NativeDicomModel/DicomAttribute[@tag=\"00100010\"]/PersonName/Alphabetic/GivenName"); + ASSERT_STREQ("Gildong", node.node().text().as_string()); + + node = SelectNode(doc, "//NativeDicomModel/DicomAttribute[@tag=\"00100010\"]/PersonName/Ideographic/FamilyName"); + ASSERT_EQ(utf8.substr(13, 3), node.node().text().as_string()); + + node = SelectNode(doc, "//NativeDicomModel/DicomAttribute[@tag=\"00100010\"]/PersonName/Ideographic/GivenName"); + ASSERT_EQ(utf8.substr(17, 6), node.node().text().as_string()); + + node = SelectNode(doc, "//NativeDicomModel/DicomAttribute[@tag=\"00100010\"]/PersonName/Phonetic/FamilyName"); + ASSERT_EQ(utf8.substr(24, 3), node.node().text().as_string()); + + node = SelectNode(doc, "//NativeDicomModel/DicomAttribute[@tag=\"00100010\"]/PersonName/Phonetic/GivenName"); + ASSERT_EQ(utf8.substr(28), node.node().text().as_string()); +#endif + + { + DicomMap m; + m.FromDicomWeb(visitor.GetResult()); + ASSERT_EQ(2u, m.GetSize()); + + std::string s; + ASSERT_TRUE(m.LookupStringValue(s, DICOM_TAG_SPECIFIC_CHARACTER_SET, false)); + ASSERT_EQ("ISO 2022 IR 149", s); + + ASSERT_TRUE(m.LookupStringValue(s, DICOM_TAG_PATIENT_NAME, false)); + std::vector<std::string> v; + Toolbox::TokenizeString(v, s, '='); + ASSERT_EQ(3u, v.size()); + ASSERT_EQ("Hong^Gildong", v[0]); + ASSERT_EQ(utf8, s); + } +} + + +TEST(Toolbox, EncodingsJapaneseKanji) +{ + // http://dicom.nema.org/MEDICAL/dicom/current/output/chtml/part05/sect_H.3.html + + std::string japanese = DecodeFromSpecification( + "05/09 06/01 06/13 06/01 06/04 06/01 05/14 05/04 06/01 07/02 06/15 07/05 03/13 " + "01/11 02/04 04/02 03/11 03/03 04/05 04/04 01/11 02/08 04/02 05/14 01/11 02/04 " + "04/02 04/02 04/00 04/15 03/10 01/11 02/08 04/02 03/13 01/11 02/04 04/02 02/04 " + "06/04 02/04 05/14 02/04 04/00 01/11 02/08 04/02 05/14 01/11 02/04 04/02 02/04 " + "03/15 02/04 06/13 02/04 02/06 01/11 02/08 04/02"); + + // This array can be re-generated using command-line: + // echo -n "Yamada^Tarou=..." | hexdump -v -e '14/1 "0x%02x, "' -e '"\n"' + static const uint8_t utf8raw[] = { + 0x59, 0x61, 0x6d, 0x61, 0x64, 0x61, 0x5e, 0x54, 0x61, 0x72, 0x6f, 0x75, 0x3d, 0xe5, + 0xb1, 0xb1, 0xe7, 0x94, 0xb0, 0x5e, 0xe5, 0xa4, 0xaa, 0xe9, 0x83, 0x8e, 0x3d, 0xe3, + 0x82, 0x84, 0xe3, 0x81, 0xbe, 0xe3, 0x81, 0xa0, 0x5e, 0xe3, 0x81, 0x9f, 0xe3, 0x82, + 0x8d, 0xe3, 0x81, 0x86 + }; + + std::string utf8(reinterpret_cast<const char*>(utf8raw), sizeof(utf8raw)); + + ParsedDicomFile dicom(false); + dicom.ReplacePlainString(DICOM_TAG_SPECIFIC_CHARACTER_SET, "\\ISO 2022 IR 87"); + ASSERT_TRUE(dicom.GetDcmtkObject().getDataset()->putAndInsertString + (DCM_PatientName, japanese.c_str(), OFBool(true)).good()); + + bool hasCodeExtensions; + Encoding encoding = dicom.DetectEncoding(hasCodeExtensions); + ASSERT_EQ(Encoding_JapaneseKanji, encoding); + ASSERT_TRUE(hasCodeExtensions); + + std::string value; + ASSERT_TRUE(dicom.GetTagValue(value, DICOM_TAG_PATIENT_NAME)); + ASSERT_EQ(utf8, value); + + DicomWebJsonVisitor visitor; + dicom.Apply(visitor); + ASSERT_EQ(utf8.substr(0, 12), visitor.GetResult()["00100010"]["Value"][0]["Alphabetic"].asString()); + ASSERT_EQ(utf8.substr(13, 13), visitor.GetResult()["00100010"]["Value"][0]["Ideographic"].asString()); + ASSERT_EQ(utf8.substr(27), visitor.GetResult()["00100010"]["Value"][0]["Phonetic"].asString()); + +#if ORTHANC_ENABLE_PUGIXML == 1 + // http://dicom.nema.org/medical/dicom/current/output/chtml/part18/sect_F.3.html#table_F.3.1-1 + std::string xml; + visitor.FormatXml(xml); + + pugi::xml_document doc; + doc.load_buffer(xml.c_str(), xml.size()); + + pugi::xpath_node node = SelectNode(doc, "//NativeDicomModel/DicomAttribute[@tag=\"00080005\"]/Value"); + ASSERT_STREQ("ISO 2022 IR 87", node.node().text().as_string()); + + node = SelectNode(doc, "//NativeDicomModel/DicomAttribute[@tag=\"00080005\"]"); + ASSERT_STREQ("CS", node.node().attribute("vr").value()); + + node = SelectNode(doc, "//NativeDicomModel/DicomAttribute[@tag=\"00100010\"]"); + ASSERT_STREQ("PN", node.node().attribute("vr").value()); + + node = SelectNode(doc, "//NativeDicomModel/DicomAttribute[@tag=\"00100010\"]/PersonName/Alphabetic/FamilyName"); + ASSERT_STREQ("Yamada", node.node().text().as_string()); + + node = SelectNode(doc, "//NativeDicomModel/DicomAttribute[@tag=\"00100010\"]/PersonName/Alphabetic/GivenName"); + ASSERT_STREQ("Tarou", node.node().text().as_string()); + + node = SelectNode(doc, "//NativeDicomModel/DicomAttribute[@tag=\"00100010\"]/PersonName/Ideographic/FamilyName"); + ASSERT_EQ(utf8.substr(13, 6), node.node().text().as_string()); + + node = SelectNode(doc, "//NativeDicomModel/DicomAttribute[@tag=\"00100010\"]/PersonName/Ideographic/GivenName"); + ASSERT_EQ(utf8.substr(20, 6), node.node().text().as_string()); + + node = SelectNode(doc, "//NativeDicomModel/DicomAttribute[@tag=\"00100010\"]/PersonName/Phonetic/FamilyName"); + ASSERT_EQ(utf8.substr(27, 9), node.node().text().as_string()); + + node = SelectNode(doc, "//NativeDicomModel/DicomAttribute[@tag=\"00100010\"]/PersonName/Phonetic/GivenName"); + ASSERT_EQ(utf8.substr(37), node.node().text().as_string()); +#endif + + { + DicomMap m; + m.FromDicomWeb(visitor.GetResult()); + ASSERT_EQ(2u, m.GetSize()); + + std::string s; + ASSERT_TRUE(m.LookupStringValue(s, DICOM_TAG_SPECIFIC_CHARACTER_SET, false)); + ASSERT_EQ("ISO 2022 IR 87", s); + + ASSERT_TRUE(m.LookupStringValue(s, DICOM_TAG_PATIENT_NAME, false)); + std::vector<std::string> v; + Toolbox::TokenizeString(v, s, '='); + ASSERT_EQ(3u, v.size()); + ASSERT_EQ("Yamada^Tarou", v[0]); + ASSERT_EQ(utf8, s); + } +} + + + +TEST(Toolbox, EncodingsChinese3) +{ + // http://dicom.nema.org/MEDICAL/dicom/current/output/chtml/part05/sect_J.3.html + + static const uint8_t chinese[] = { + 0x57, 0x61, 0x6e, 0x67, 0x5e, 0x58, 0x69, 0x61, 0x6f, 0x44, 0x6f, + 0x6e, 0x67, 0x3d, 0xcd, 0xf5, 0x5e, 0xd0, 0xa1, 0xb6, 0xab, 0x3d, 0x00 + }; + + ParsedDicomFile dicom(false); + dicom.ReplacePlainString(DICOM_TAG_SPECIFIC_CHARACTER_SET, "GB18030"); + ASSERT_TRUE(dicom.GetDcmtkObject().getDataset()->putAndInsertString + (DCM_PatientName, reinterpret_cast<const char*>(chinese), OFBool(true)).good()); + + bool hasCodeExtensions; + Encoding encoding = dicom.DetectEncoding(hasCodeExtensions); + ASSERT_EQ(Encoding_Chinese, encoding); + ASSERT_FALSE(hasCodeExtensions); + + std::string value; + ASSERT_TRUE(dicom.GetTagValue(value, DICOM_TAG_PATIENT_NAME)); + + std::vector<std::string> tokens; + Orthanc::Toolbox::TokenizeString(tokens, value, '='); + ASSERT_EQ(3u, tokens.size()); + ASSERT_EQ("Wang^XiaoDong", tokens[0]); + ASSERT_TRUE(tokens[2].empty()); + + std::vector<std::string> middle; + Orthanc::Toolbox::TokenizeString(middle, tokens[1], '^'); + ASSERT_EQ(2u, middle.size()); + ASSERT_EQ(3u, middle[0].size()); + ASSERT_EQ(6u, middle[1].size()); + + // CDF5 in GB18030 + ASSERT_EQ(static_cast<char>(0xe7), middle[0][0]); + ASSERT_EQ(static_cast<char>(0x8e), middle[0][1]); + ASSERT_EQ(static_cast<char>(0x8b), middle[0][2]); + + // D0A1 in GB18030 + ASSERT_EQ(static_cast<char>(0xe5), middle[1][0]); + ASSERT_EQ(static_cast<char>(0xb0), middle[1][1]); + ASSERT_EQ(static_cast<char>(0x8f), middle[1][2]); + + // B6AB in GB18030 + ASSERT_EQ(static_cast<char>(0xe4), middle[1][3]); + ASSERT_EQ(static_cast<char>(0xb8), middle[1][4]); + ASSERT_EQ(static_cast<char>(0x9c), middle[1][5]); +} + + +TEST(Toolbox, EncodingsChinese4) +{ + // http://dicom.nema.org/MEDICAL/dicom/current/output/chtml/part05/sect_J.4.html + + static const uint8_t chinese[] = { + 0x54, 0x68, 0x65, 0x20, 0x66, 0x69, 0x72, 0x73, 0x74, 0x20, 0x6c, 0x69, 0x6e, + 0x65, 0x20, 0x69, 0x6e, 0x63, 0x6c, 0x75, 0x64, 0x65, 0x73, 0xd6, 0xd0, 0xce, + 0xc4, 0x2e, 0x0d, 0x0a, 0x54, 0x68, 0x65, 0x20, 0x73, 0x65, 0x63, 0x6f, 0x6e, + 0x64, 0x20, 0x6c, 0x69, 0x6e, 0x65, 0x20, 0x69, 0x6e, 0x63, 0x6c, 0x75, 0x64, + 0x65, 0x73, 0xd6, 0xd0, 0xce, 0xc4, 0x2c, 0x20, 0x74, 0x6f, 0x6f, 0x2e, 0x0d, + 0x0a, 0x54, 0x68, 0x65, 0x20, 0x74, 0x68, 0x69, 0x72, 0x64, 0x20, 0x6c, 0x69, + 0x6e, 0x65, 0x2e, 0x0d, 0x0a, 0x00 + }; + + static const uint8_t patternRaw[] = { + 0xe4, 0xb8, 0xad, 0xe6, 0x96, 0x87 + }; + + const std::string pattern(reinterpret_cast<const char*>(patternRaw), sizeof(patternRaw)); + + ParsedDicomFile dicom(false); + dicom.ReplacePlainString(DICOM_TAG_SPECIFIC_CHARACTER_SET, "GB18030"); + ASSERT_TRUE(dicom.GetDcmtkObject().getDataset()->putAndInsertString + (DCM_PatientComments, reinterpret_cast<const char*>(chinese), OFBool(true)).good()); + + bool hasCodeExtensions; + Encoding encoding = dicom.DetectEncoding(hasCodeExtensions); + ASSERT_EQ(Encoding_Chinese, encoding); + ASSERT_FALSE(hasCodeExtensions); + + std::string value; + ASSERT_TRUE(dicom.GetTagValue(value, DICOM_TAG_PATIENT_COMMENTS)); + + std::vector<std::string> lines; + Orthanc::Toolbox::TokenizeString(lines, value, '\n'); + ASSERT_EQ(4u, lines.size()); + ASSERT_TRUE(boost::starts_with(lines[0], "The first line includes")); + ASSERT_TRUE(boost::ends_with(lines[0], ".\r")); + ASSERT_TRUE(lines[0].find(pattern) != std::string::npos); + ASSERT_TRUE(boost::starts_with(lines[1], "The second line includes")); + ASSERT_TRUE(boost::ends_with(lines[1], ", too.\r")); + ASSERT_TRUE(lines[1].find(pattern) != std::string::npos); + ASSERT_EQ("The third line.\r", lines[2]); + ASSERT_FALSE(lines[1].find(pattern) == std::string::npos); + ASSERT_TRUE(lines[3].empty()); +} + + +TEST(Toolbox, EncodingsSimplifiedChinese2) +{ + // http://dicom.nema.org/MEDICAL/dicom/current/output/chtml/part05/sect_K.2.html + + static const uint8_t chinese[] = { + 0x5a, 0x68, 0x61, 0x6e, 0x67, 0x5e, 0x58, 0x69, 0x61, 0x6f, 0x44, 0x6f, + 0x6e, 0x67, 0x3d, 0x1b, 0x24, 0x29, 0x41, 0xd5, 0xc5, 0x5e, 0x1b, 0x24, + 0x29, 0x41, 0xd0, 0xa1, 0xb6, 0xab, 0x3d, 0x20, 0x00 + }; + + // echo -n "Zhang^XiaoDong=..." | hexdump -v -e '14/1 "0x%02x, "' -e '"\n"' + static const uint8_t utf8[] = { + 0x5a, 0x68, 0x61, 0x6e, 0x67, 0x5e, 0x58, 0x69, 0x61, 0x6f, 0x44, 0x6f, 0x6e, 0x67, + 0x3d, 0xe5, 0xbc, 0xa0, 0x5e, 0xe5, 0xb0, 0x8f, 0xe4, 0xb8, 0x9c, 0x3d + }; + + ParsedDicomFile dicom(false); + dicom.ReplacePlainString(DICOM_TAG_SPECIFIC_CHARACTER_SET, "\\ISO 2022 IR 58"); + ASSERT_TRUE(dicom.GetDcmtkObject().getDataset()->putAndInsertString + (DCM_PatientName, reinterpret_cast<const char*>(chinese), OFBool(true)).good()); + + bool hasCodeExtensions; + Encoding encoding = dicom.DetectEncoding(hasCodeExtensions); + ASSERT_EQ(Encoding_SimplifiedChinese, encoding); + ASSERT_TRUE(hasCodeExtensions); + + std::string value; + ASSERT_TRUE(dicom.GetTagValue(value, DICOM_TAG_PATIENT_NAME)); + ASSERT_EQ(value, std::string(reinterpret_cast<const char*>(utf8), sizeof(utf8))); +} + + +TEST(Toolbox, EncodingsSimplifiedChinese3) +{ + // http://dicom.nema.org/MEDICAL/dicom/current/output/chtml/part05/sect_K.2.html + + static const uint8_t chinese[] = { + 0x31, 0x2e, 0x1b, 0x24, 0x29, 0x41, 0xb5, 0xda, 0xd2, 0xbb, 0xd0, 0xd0, 0xce, 0xc4, 0xd7, 0xd6, 0xa1, 0xa3, 0x0d, 0x0a, + 0x32, 0x2e, 0x1b, 0x24, 0x29, 0x41, 0xb5, 0xda, 0xb6, 0xfe, 0xd0, 0xd0, 0xce, 0xc4, 0xd7, 0xd6, 0xa1, 0xa3, 0x0d, 0x0a, + 0x33, 0x2e, 0x1b, 0x24, 0x29, 0x41, 0xb5, 0xda, 0xc8, 0xfd, 0xd0, 0xd0, 0xce, 0xc4, 0xd7, 0xd6, 0xa1, 0xa3, 0x0d, 0x0a, 0x00 + }; + + static const uint8_t line1[] = { + 0x31, 0x2e, 0xe7, 0xac, 0xac, 0xe4, 0xb8, 0x80, 0xe8, 0xa1, 0x8c, 0xe6, 0x96, 0x87, + 0xe5, 0xad, 0x97, 0xe3, 0x80, 0x82, '\r' + }; + + static const uint8_t line2[] = { + 0x32, 0x2e, 0xe7, 0xac, 0xac, 0xe4, 0xba, 0x8c, 0xe8, 0xa1, 0x8c, 0xe6, 0x96, 0x87, + 0xe5, 0xad, 0x97, 0xe3, 0x80, 0x82, '\r' + }; + + static const uint8_t line3[] = { + 0x33, 0x2e, 0xe7, 0xac, 0xac, 0xe4, 0xb8, 0x89, 0xe8, 0xa1, 0x8c, 0xe6, 0x96, 0x87, + 0xe5, 0xad, 0x97, 0xe3, 0x80, 0x82, '\r' + }; + + ParsedDicomFile dicom(false); + dicom.ReplacePlainString(DICOM_TAG_SPECIFIC_CHARACTER_SET, "\\ISO 2022 IR 58"); + ASSERT_TRUE(dicom.GetDcmtkObject().getDataset()->putAndInsertString + (DCM_PatientName, reinterpret_cast<const char*>(chinese), OFBool(true)).good()); + + bool hasCodeExtensions; + Encoding encoding = dicom.DetectEncoding(hasCodeExtensions); + ASSERT_EQ(Encoding_SimplifiedChinese, encoding); + ASSERT_TRUE(hasCodeExtensions); + + std::string value; + ASSERT_TRUE(dicom.GetTagValue(value, DICOM_TAG_PATIENT_NAME)); + + std::vector<std::string> lines; + Toolbox::TokenizeString(lines, value, '\n'); + ASSERT_EQ(4u, lines.size()); + ASSERT_EQ(std::string(reinterpret_cast<const char*>(line1), sizeof(line1)), lines[0]); + ASSERT_EQ(std::string(reinterpret_cast<const char*>(line2), sizeof(line2)), lines[1]); + ASSERT_EQ(std::string(reinterpret_cast<const char*>(line3), sizeof(line3)), lines[2]); + ASSERT_TRUE(lines[3].empty()); +} + + + + +#if ORTHANC_ENABLE_DCMTK_TRANSCODING == 1 + +#include "../Sources/DicomNetworking/DicomStoreUserConnection.h" +#include "../Sources/DicomParsing/DcmtkTranscoder.h" + +TEST(Toto, DISABLED_Transcode3) +{ + DicomAssociationParameters p; + p.SetRemotePort(2000); + + DicomStoreUserConnection scu(p); + scu.SetCommonClassesProposed(false); + scu.SetRetiredBigEndianProposed(true); + + DcmtkTranscoder transcoder; + + for (int j = 0; j < 2; j++) + for (int i = 0; i <= DicomTransferSyntax_XML; i++) + { + DicomTransferSyntax a = (DicomTransferSyntax) i; + + std::string path = ("/home/jodogne/Subversion/orthanc-tests/Database/TransferSyntaxes/" + + std::string(GetTransferSyntaxUid(a)) + ".dcm"); + if (Orthanc::SystemToolbox::IsRegularFile(path)) + { + printf("\n======= %s\n", GetTransferSyntaxUid(a)); + + std::string source; + Orthanc::SystemToolbox::ReadFile(source, path); + + std::string c, i; + try + { + scu.Transcode(c, i, transcoder, source.c_str(), source.size(), false, "", 0); + } + catch (OrthancException& e) + { + if (e.GetErrorCode() == ErrorCode_NotImplemented) + { + LOG(ERROR) << "cannot transcode " << GetTransferSyntaxUid(a); + } + else + { + throw e; + } + } + } + } +} + + +TEST(Toto, DISABLED_Transcode4) +{ + std::string source; + Orthanc::SystemToolbox::ReadFile(source, "/home/jodogne/Subversion/orthanc-tests/Database/KarstenHilbertRF.dcm"); + + std::unique_ptr<DcmFileFormat> toto(FromDcmtkBridge::LoadFromMemoryBuffer(source.c_str(), source.size())); + const std::string sourceUid = IDicomTranscoder::GetSopInstanceUid(*toto); + + DicomTransferSyntax sourceSyntax; + ASSERT_TRUE(FromDcmtkBridge::LookupOrthancTransferSyntax(sourceSyntax, *toto)); + + DcmtkTranscoder transcoder; + + for (int i = 0; i <= DicomTransferSyntax_XML; i++) + { + DicomTransferSyntax a = (DicomTransferSyntax) i; + + std::set<DicomTransferSyntax> s; + s.insert(a); + + std::string t; + + IDicomTranscoder::DicomImage source, target; + source.AcquireParsed(dynamic_cast<DcmFileFormat*>(toto->clone())); + + if (!transcoder.Transcode(target, source, s, true)) + { + printf("**************** CANNOT: [%s] => [%s]\n", + GetTransferSyntaxUid(sourceSyntax), GetTransferSyntaxUid(a)); + } + else + { + DicomTransferSyntax targetSyntax; + ASSERT_TRUE(FromDcmtkBridge::LookupOrthancTransferSyntax(targetSyntax, target.GetParsed())); + + ASSERT_EQ(targetSyntax, a); + bool lossy = (a == DicomTransferSyntax_JPEGProcess1 || + a == DicomTransferSyntax_JPEGProcess2_4 || + a == DicomTransferSyntax_JPEGLSLossy); + + printf("SIZE: %lu\n", t.size()); + if (sourceUid == IDicomTranscoder::GetSopInstanceUid(target.GetParsed())) + { + ASSERT_FALSE(lossy); + } + else + { + ASSERT_TRUE(lossy); + } + } + } +} + +#endif