Set pixel value of 16 bit grayscale QImage

Question

I have an 16 bit image of width ("imagewidth") and height ("imageheight"). The data is currently stored in an unsigned short int array of length ("imagewidth"*"imageheight")

I want to create an 16 bit grayscale QImage (using Qt 5.14) from my dataset which is called "data".

This is the code that I am using:

QImage image = Qimage(imagewidth,imageheight,QImage::Format_Grayscale16);

for(int i=0;i<imagewidth;i++)
{
   for(int j=0;j<imageheight;j++)
   {

    uint pixelval = data[i+j*imagewidth];
    QRgb color = qRgb(pixelval, pixelval, pixelval);
    image.setPixel(i,j, color);

   }
}

The code is working and I am getting an image but I only get values in increments of 255... So 0, 255, ...

How I can set the actual pixel value for each pixel from 0 to 65535 ?

Answer 1

QRgba64 is the right choice for 16 bit (per component) colors.

Another option is to retrieve color with QImage::pixelColor() (and setting with QImage::setPixelColor()) which should be more or less depth agnostic.

Function qRgb() is a bad choice because it deals with 8 bit (per component) colors by intention.

From Qt doc.:

QRgb QColor::qRgb(int r, int g, int b)

Returns the ARGB quadruplet (255, r, g, b).

The value 255 for alpha gives a first hint but inspecting the result type QRgb makes this obvious:

typedef QColor::QRgb

An ARGB quadruplet on the format #AARRGGBB, equivalent to an unsigned int.

A look into source code on woboq.org supports this:

inline Q_DECL_CONSTEXPR QRgb qRgb(int r, int g, int b)// set RGB value
{ return (0xffu << 24) | ((r & 0xffu) << 16) | ((g & 0xffu) << 8) | (b & 0xffu); }

A small sample to illustrate this:

#include <QtWidgets>

int main(int argc, char **argv)
{
  qDebug() << "Qt Version:" << QT_VERSION_STR;
  qDebug() << "qRgb(1 << 12, 1 << 12, 1 << 12):"
    << hex << qRgb(1 << 12, 1 << 12, 1 << 12);
  qDebug() << "QColor(qRgb(1 << 12, 1 << 12, 1 << 12)):"
    << QColor(qRgb(1 << 12, 1 << 12, 1 << 12));
  qDebug() << "QColor().fromRgba64(1 << 12, 1 << 12, 1 << 12):"
    << QColor().fromRgba64(1 << 12, 1 << 12, 1 << 12);
  // done
  return 0;
}

Output:

Qt Version: 5.11.2
qRgb(1 << 12, 1 << 12, 1 << 12): ff000000
QColor(qRgb(1 << 12, 1 << 12, 1 << 12)): QColor(ARGB 1, 0, 0, 0)
QColor().fromRgba64(1 << 12, 1 << 12, 1 << 12): QColor(ARGB 1, 0.062501, 0.062501, 0.062501)

---

An alternative to using QColor and companions is to write the data values directly into image:

QImage image = Qimage(imagewidth, imageheight, QImage::Format_Grayscale16);
for (int j = 0; j < imageheight; ++j) {
  quint16 *dst = (quint16*)(image.bits() + j * image.bytesPerLine());
  for (int i = 0; i < imagewidth; ++i) {
    dst[i] = data[i + j * imagewidth];
  }
}

This is surely faster and more accurate than converting data value to a color which is converted to a gray level again.

Please note, that I swapped the loops for rows and columns. Processing consecutive bytes in source (data) and destination (dst) will improve cache locality and pay off in speed.

---

Images with 16 bit depth are rather new in Qt at the time of writing.

New color formats with 16 bit depth per component were added in Qt 5.12:

• QImage::Format_RGBX64 = 25
  The image is stored using a 64-bit halfword-ordered RGB(x) format (16-16-16-16). This is the same as the Format_RGBX64 except alpha must always be 65535. (added in Qt 5.12)
• QImage::Format_RGBA64 = 26
  The image is stored using a 64-bit halfword-ordered RGBA format (16-16-16-16). (added in Qt 5.12)
• QImage::Format_RGBA64_Premultiplied = 27
  The image is stored using a premultiplied 64-bit halfword-ordered RGBA format (16-16-16-16). (added in Qt 5.12)

The grayscale with 16 bit depth was added in Qt 5.13:

• QImage::Format_Grayscale16 = 28
  The image is stored using an 16-bit grayscale format. (added in Qt 5.13)

(Doc. copied from enum QImage::Format)

To fiddle a bit with this, I made a sample application for conversion of 16 bit-per-component RGB image to 16 bit grayscale.

testQImageGray16.cc:

#include <QtWidgets>

QImage imageToGray16(const QImage &qImg)
{
  QImage qImgGray(qImg.width(), qImg.height(), QImage::Format_Grayscale16);
  for (int y = 0; y < qImg.height(); ++y) {
    for (int x = 0; x < qImg.width(); ++x) {
      qImgGray.setPixelColor(x, y, qImg.pixelColor(x, y));
    }
  }
  return qImgGray;
}

class Canvas: public QWidget {
  private:
    QImage _qImg;
  public:
    std::function<void(QPoint)> sigMouseMove;

  public:
    Canvas(QWidget *pQParent = nullptr):
      QWidget(pQParent)
    {
      setMouseTracking(true);
    }
    Canvas(const QImage &qImg, QWidget *pQParent = nullptr):
      QWidget(pQParent), _qImg(qImg)
    {
      setMouseTracking(true);
    }
    virtual ~Canvas() = default;
    Canvas(const Canvas&) = delete;
    Canvas& operator=(const Canvas&) = delete;

  public:
    virtual QSize sizeHint() const { return _qImg.size(); }
    const QImage& image() const { return _qImg; }
    void setImage(const QImage &qImg) { _qImg = qImg; update(); }
  protected:
    virtual void paintEvent(QPaintEvent *pQEvent) override;
    virtual void mouseMoveEvent(QMouseEvent *pQEvent) override;
};

void Canvas::paintEvent(QPaintEvent *pQEvent)
{
  QWidget::paintEvent(pQEvent);
  QPainter qPainter(this);
  qPainter.drawImage(0, 0, _qImg);
}

void Canvas::mouseMoveEvent(QMouseEvent *pQEvent)
{
  if (sigMouseMove) sigMouseMove(pQEvent->pos());
}

QString getInfo(const QImage &qImg)
{
  QString qStr;
  QDebug(&qStr) << "Image Info:\n" << qImg;
  for (int i = 0, len = qStr.length(); i < qStr.length(); ++i) {
    if (qStr[i] == ',' && i + 1 < len && qStr[i + 1] != ' ') qStr[i] = '\n';
  }
  return qStr;
}

QString getPixelInfo(const QImage &qImg, QPoint pos)
{
  if (!QRect(QPoint(0, 0), qImg.size()).contains(pos)) return QString();
  const int bytes = (qImg.depth() + 7) / 8; assert(bytes > 0);
  const QByteArray data(
    (const char*)(qImg.bits()
      + pos.y() * qImg.bytesPerLine()
      + (pos.x() * qImg.depth() + 7) / 8),
    bytes);
  QString qStr;
  QDebug(&qStr) << pos << ":" << qImg.pixelColor(pos)
    << "raw:" << QString("#") + data.toHex();
  return qStr;
}

int main(int argc, char **argv)
{
  qDebug() << "Qt Version:" << QT_VERSION_STR;
  QApplication app(argc, argv);
  // load sample data
  const QImage qImgRGB16("pnggrad16rgb.png"/*, QImage::Format_RGBX64*/);
  // setup GUI
  QWidget winMain;
  QGridLayout qGrid;
  int col = 0, row = 0;
  QLabel qLblRGBInfo(getInfo(qImgRGB16));
  qGrid.addWidget(&qLblRGBInfo, row++, col);
  Canvas qCanvasRGB(qImgRGB16);
  qGrid.addWidget(&qCanvasRGB, row++, col);
  QLabel qLblRGB;
  qGrid.addWidget(&qLblRGB, row++, col);
  row = 0; ++col;
  Canvas qCanvasGray(qImgRGB16.convertToFormat(QImage::Format_Grayscale16));
  QLabel qLblGrayInfo;
  qGrid.addWidget(&qLblGrayInfo, row++, col);
  qGrid.addWidget(&qCanvasGray, row++, col);
  QLabel qLblGray;
  qGrid.addWidget(&qLblGray, row++, col);
  QHBoxLayout qHBoxQImageConvert;
  QButtonGroup qBtnGrpQImageConvert;
  QRadioButton qTglQImageConvertBuiltIn("Use QImage::convertToFormat()");
  qBtnGrpQImageConvert.addButton(&qTglQImageConvertBuiltIn);
  qTglQImageConvertBuiltIn.setChecked(true);
  qHBoxQImageConvert.addWidget(&qTglQImageConvertBuiltIn);
  QRadioButton qTglQImageConvertCustom("Use imageToGray16()");
  qBtnGrpQImageConvert.addButton(&qTglQImageConvertCustom);
  qHBoxQImageConvert.addWidget(&qTglQImageConvertCustom);
  qGrid.addLayout(&qHBoxQImageConvert, row++, col);
  winMain.setLayout(&qGrid);
  winMain.show();
  // install signal handlers
  auto updatePixelInfo = [&](QPoint pos)
  {
    qLblRGB.setText (getPixelInfo(qCanvasRGB.image(), pos));
    qLblGray.setText(getPixelInfo(qCanvasGray.image(), pos));
  };
  qCanvasRGB.sigMouseMove = updatePixelInfo;
  qCanvasGray.sigMouseMove = updatePixelInfo;
  auto updateGrayImage = [&](bool customConvert)
  {
    qCanvasGray.setImage(customConvert
      ? qImgRGB16.convertToFormat(QImage::Format_Grayscale16)
      : imageToGray16(qImgRGB16));
    qLblGrayInfo.setText(getInfo(qCanvasGray.image()));
    qLblGray.setText(QString());
  };
  QObject::connect(&qTglQImageConvertBuiltIn, &QRadioButton::toggled,
    [&](bool checked) { if (checked) updateGrayImage(false); });
  QObject::connect(&qTglQImageConvertCustom, &QRadioButton::toggled,
    [&](bool checked) { if (checked) updateGrayImage(true); });
  // runtime loop
  updateGrayImage(false);
  return app.exec();
}

and a build-script for CMake CMakeLists.txt:

project(QImageGray16)

cmake_minimum_required(VERSION 3.10.0)

set_property(GLOBAL PROPERTY USE_FOLDERS ON)
set(CMAKE_CXX_STANDARD 17)
set(CMAKE_CXX_STANDARD_REQUIRED ON)
set(CMAKE_CXX_EXTENSIONS OFF)

find_package(Qt5Widgets CONFIG REQUIRED)

include_directories("${CMAKE_SOURCE_DIR}")

add_executable(testQImageGray16
  testQImageGray16.cc)

target_link_libraries(testQImageGray16
  Qt5::Widgets)

# define QT_NO_KEYWORDS to prevent confusion between of Qt signal-slots and
# other signal-slot APIs
target_compile_definitions(testQImageGray16 PUBLIC QT_NO_KEYWORDS)

I downloaded a sample image pnggrad16rgb.png from www.fnordware.com/superpng/samples.html. (Other sample images can be found on The "official" test-suite for PNG.)

After building and running in VS2017, I got the following snapshot:

When mouse is moved over the displayed images, the bottom label shows the current position in image and the respective pixel as QColor and raw hex-values.

Out of curiosity, I implemented the approach of OP with two nested loops (fixing the qRgb() issue):

QImage imageToGray16(const QImage &qImg)
{
  QImage qImgGray(qImg.width(), qImg.height(), QImage::Format_Grayscale16);
  for (int y = 0; y < qImg.height(); ++y) {
    for (int x = 0; x < qImg.width(); ++x) {
      qImgGray.setPixelColor(x, y, qImg.pixelColor(x, y));
    }
  }
  return qImgGray;
}

to compare the result with the QImage::convertToFormat() which I found in the Qt doc.

QImage QImage::convertToFormat(QImage::Format format, Qt::ImageConversionFlags flags = Qt::AutoColor) const &

QImage QImage::convertToFormat(QImage::Format format, Qt::ImageConversionFlags flags = Qt::AutoColor) &&

Returns a copy of the image in the given format.

The specified image conversion flags control how the image data is handled during the conversion process.

That's interesting:

There is no visible difference. Though, this might not be a surprise considering that the display may reduce color depth to 8 or 10 (beside of the fact that humans are probably not able to differentiate 2¹⁶ shades of gray nor 2¹⁶³ RGB values.

However wiping over the images, I realized that with QImage::convertToFormat(), the first and second byte of each pixel were always identical (e.g. b2b2), which is not the case when custom conversion with imageToGray16() is used.

I didn't dig deeper but it could be worth a further investigation which of these methods is actually more accurate.

Answer 2

@Scheff, thanks for the input ! I am now using the following code:

Qimage = QImage(imagewidth,imageheight,QImage::Format_Grayscale16);


    for (int j = 0; j < imageheight; ++j)
    {
       quint16 *dst =  reinterpret_cast<quint16*>(Qimage.bits() + j * Qimage.bytesPerLine());

       for (int i = 0; i < imagewidth; ++i)
       {

            unsigned short pixelval =  static_cast<unsigned short>(image[i + j * imagewidth]);

            dst[i] = pixelval;

       }

    }