OpenCV - is there anything like delete text? - c++

I need such thing because it seems to me that when I do
cvRectangle( CVframe, UL, LR, CV_RGB(0,256,53), CV_FILLED);
string cvtext;
cvtext += timeStr;
cvPutText(CVframe, cvtext.c_str(), cvPoint(0,(h/2+10)), &font , CV_RGB(0,0,0));
each time 24 times per second cvRectangle does not overlay old text...

There's no built-in cvDeleteText or anything like that, and probably for good reason. Whenever you put text on an image, it overwrites the pixels in that image, just as if you had set their values to CV_RGB(0,0,0) individually. If you wanted to undo that operation, you'd need to store whatever had been there beforehand. Since not everyone wants to do that, it would be a waste of space and time if cvPutText automatically kept track of the pixels it wrote over.
Probably the best approach would be to have two frames, one of which is never touched by text. The code would look something like this.
//Initializing, before your loop that executes 24 times per second:
CvArr *CVframe, *CVframeWithText; // make sure they're the same size and format
while (looping) {
cvRectangle( CVframe, UL, LR, CV_RGB(0,256,53), CV_FILLED);
// And anything else non-text-related, do it to CVframe.
// Now we want to copy the frame without text.
cvCopy(CVframe, CVframeWithText);
string cvtext;
cvtext += timeStr;
// And now, notice in the following line that
// we're not overwriting any pixels in CVframe
cvPutText(CVframeWithText, cvtext.c_str(), cvPoint(0,(h/2+10)), &font , CV_RGB(0,0,0));
// And then display CVframeWithText.
// Now, the contents of CVframe are the same as if we'd "deleted" the text;
// in fact, we never wrote text to CVframe in the first place.
Hope this helps!

I don't know, but maybe this can help.
Make two mouse click events, LeftClick and Right Click.
In my example, when I left-click the mouse it gives the pixel coordinate of the image, and when I right-click the mouse, it loads the image from the fresh. It's kindof removing what cv2.putText function placed there.
just import any image, I have used 'lena_color.tiff' as an example.
import cv2
def click_event(event, x, y, flags, params):
global img
if event == cv2.EVENT_LBUTTONDOWN:
print(x, ' ', y)
font = cv2.FONT_HERSHEY_SIMPLEX; fontSize = 2
point = '.'; text = ' (' + str(x) + ', ' + str(y) + ')'
color = (255, 0, 0); thickness=6
cv2.putText(img, point, (x,y), font, fontSize, color, thickness)
cv2.putText(img, text, (x,y), font, 0.5, color, 1)
cv2.imshow('image', img)
if event==cv2.EVENT_RBUTTONDOWN:
img = cv2.imread('lena_color.tiff', 1)
cv2.imshow('image', img)
cv2.setMouseCallback('image', click_event)
img = cv2.imread('lena_color.tiff', 1)
cv2.imshow('image', img)
cv2.setMouseCallback('image', click_event)
cv2.waitKey(0)
cv2.destroyAllWindows()

Related

Qt text size in points

I'm trying to print an invoice document on A4 in millimetres instead of the default device units. Except that when changing the units to millimetres the pointsize of text on the printed document no longer matches up with the pointsize in for instance Word or Adobe Illustrator. I tried converting the point size to the corresponding pixel size, but they had issues.
QFont::SetPixelSize only takes an int so if the calculations falls below 1 it will trunctuate to 0
font.setPixelSize((9.0 * 72.0) / printer.resolution());
And the other method made the text the correct vertical size, but there are some artefacts:
int phys_w = printer.width();
font.setPointSizeF((9.0 / phys_w) * 210.0);
Where you can see the unusually large gaps between some characters. (Perhaps there is some of precision issue inside Qt its text drawing code?)
Here is a minimal example showing the issues:
QPrinter printer(QPrinter::HighResolution);
printer.setPageSize(QPrinter::A4);
printer.setOrientation(QPrinter::Portrait);
printer.setFullPage(true);
printer.setPageMargins(QMarginsF(0, 0, 0, 0));
printer.setOutputFormat(QPrinter::PdfFormat);
printer.setOutputFileName("invoice.pdf");
QPainter painter(&printer);
auto page_size = printer.pageRect(QPrinter::Unit::Millimeter);
painter.setWindow(page_size.toRect());
QFont font = painter.font();
// Either this
font.setPixelSize((9.0 * 72.0) / printer.resolution());
// or this
int phys_w = printer.width();
font.setPointSizeF((9.0 / phys_w) * 210.0);
painter.setFont(font);
painter.drawText(35, 46, "John Doe");
How can I have the positioning in Millimetres (or any arbitrary unit) and have the text size be correct in points (or some correct recalculation)?
This is on Qt 5.10.0 and Windows 10.
EDIT
In the end I opted to go for a 10x scale increase (so tenths of a millimetre) which fixed the kerning issues visible for setPointSizeF. Now the last issue I'm having with the scale is that of setting the width of a line and other shapes (QPen::setWidth) and I cant find a calculation so it's in millimetres.
EDIT
In the end the linewidth didn't need any recalculations. The final code is below:
QPrinter printer(QPrinter::HighResolution);
printer.setPageSize(QPrinter::A4);
printer.setOrientation(QPrinter::Portrait);
printer.setFullPage(true);
printer.setPageMargins(QMarginsF(0, 0, 0, 0));
printer.setOutputFormat(QPrinter::NativeFormat);
QPainter painter(&printer);
painter.setWindow(0, 0, 2100, 2970);
painter.setViewport(0, 0, printer.width(), printer.height());
QFont font(fontFamily, 0, weight, italic);
font.setPointSizeF(static_cast<float>(pixelSize) / printer.width() * 2100);
I think you're dividing where you should multiply and vice versa. Take a look with the units written out explicitly:
9 points * (1 inch / 72 points) * (printer.res pixels/inch)
multiplying the units, the numerator gets (points * inch * pixels) , denominator gets (points * inch) . Cancel out the like units and you get pixels in the numerator. So the calculation should be:
font.setPixelSize(9.0 / 72.0 * printer.resolution());
For your second problem,
QPen::setWidthF(w*printer.resolution()/25.4);
where w is your desired width in mm.
By making detailed observations (printing a long sentence) it's noticed that these gaps between characters are directly related to characters themselves, wide characters and Capital letters eat up the gap (J, s , r, c) while narrow characters leave more gap (i , l , t) .. this is just to say its not a random behavior.
In general this is known as kerning nicely explained here. To minimize this, Qt sets QFont kerning to true (default) QFont Kerning ... but the problem is that kerning is much dependent on the used font and pixel, and Qt kerning enabled sometimes has not effect as in this post, probably because setting Kerning to true
(Will apply kerning between adjacent glyphs. Note that OpenType GPOS
based kerning is currently not supported QRawFont::LayoutFlags
Which means that some font Ascent / Descent will still cause limitation how the gap is controlled.
Two solutions are considered below, the first is with still sticking to default painter font, yet you can stretch the characters to enforce reasonable gab:
font.setStretch(70); // value is experimental
I don't think this is a robust solution, unless one is limited to use specific font, while font itself is not important for invoice printing, the second attractive approach is to find out best font that meets requirements: render well with set resolution / little loss when drawn to PDF (from Qt) / and most important efficient for kerning!
"MS Gothic" (ttf) for example performs well for such setup, here how it performs (without stretch)
painter.setFont(QFont("MS Gothic"));
QFont font = painter.font();
int phys_w = printer.width();
font.setPointSizeF((9.0 / phys_w) * 210.0);
//font.setStretch(70);
painter.setFont(font);
The good thing about selecting a suitable font is that more control is possible (especially for invoice printing where small space is valuable).
For example the same can printed with less space by reducing the gap between letters:
font.setLetterSpacing(QFont::PercentageSpacing,65); // 65% gap of default
And one can go even for lower font size without loosing visual clearance.
sharing below some work while trying to clarify how printing small font size is affecting gap between letters. the idea is to draw each character separately inside QRect and at the same time draw equivalent QRectF on same area.
That's to see when font size is large QRectF are drawn fit next to each other, so also each character ... while when font size goes low the adjacent QRectFs will severely overlap and gaps between characters start to get disordered.
QPen pen=painter.pen();
pen.setWidth(0.1);
painter.setPen(pen);
QString td("John Doe");
auto spacer = font.pointSizeF(); // font size used to set width of QRect of each character.
spacer *=30.0; // large Rect width.
auto b = 35.0;
for (int i=0; i < td.length() ;i++ )
{
QRectF rectf(b+=spacer,47.0,4.0,4.0);
QRect rect(b, 47.0,4.0,4.0);
QString ch = td.at(i);
//painter.drawText(b+=spacer,46,ch);
painter.drawText(rect,Qt::AlignCenter,ch);
painter.drawRect(rectf);
}
painter.end();
Result for large font size:
Next make QRectF overlap:
spacer *=10.0;
Result, letters get less gaps and wide adjacent characters get narrow gap.

Direct2D CreateTextLayout() - How to get caret coordinates

I am rendering Text using Direct2D starting with a text Layout
HRESULT hr = m_spWriteFactory->CreateTextLayout(
m_wsText.c_str( ),
m_wsText.length( ),
m_spWriteTextFormat.Get( ),
m_rect.right - m_rect.left - m_spacing.right - m_spacing.left,
m_rect.bottom - m_rect.top - m_spacing.top - m_spacing.bottom,
&m_spTextLayout
);
and then rendering it to a bitmap which I later use with Direct3D
m_sp2DDeviceContext->DrawTextLayout(
D2D1::Point2F( m_spacing.left, m_spacing.top ),
m_spTextLayout.Get( ),
m_spTextBrush.Get( )
);
I would like to draw a simple thin flashing line as a caret. I know how to draw a line and how to make it appear / disappear.
Question: How do I get the starting point and the end point coordinates for my caret line?
Simplification: If it is much easier to assume that the text consists of one line only, then that's ok. But of course a more general solution is appreciated.
Use IDWriteTextLayout's hit-testing functions to determine these:
HitTestTextPosition for mapping a text position index (relative to the first character) to a rectangle.
HitTestTextRange for getting a whole range of rectangles such as for selection.
HitTestPoint for mapping a mouse coordinate to a text position index.
For carets, this below works for all horizontal reading directions and proportional/monospace fonts:
...
DWRITE_HIT_TEST_METRICS hitTestMetrics;
float caretX, caretY;
bool isTrailingHit = false; // Use the leading character edge for simplicity here.
// Map text position index to caret coordinate and hit-test rectangle.
textLayout->HitTestTextPosition(
textPosition,
isTrailingHit,
OUT &caretX,
OUT &caretY,
OUT &hitTestMetrics
);
// Respect user settings.
DWORD caretWidth = 1;
SystemParametersInfo(SPI_GETCARETWIDTH, 0, OUT &caretWidth, 0);
DWORD halfCaretWidth = caretWidth / 2u;
// Draw a thin rectangle.
D2D1::RectF caretRect = {
layoutOriginX + caretX - halfCaretWidth,
layoutOriginY + hitTestMetrics.top,
layoutOriginX + caretX + (caretWidth - halfCaretWidth),
layoutOriginY + hitTestMetrics.top + hitTestMetrics.height
};
solidColorBrush->SetColor(D2D1::ColorF::AliceBlue);
d2dRenderTarget->FillRectangle(&caretRect, solidColorBrush);
Notes:
The above code as-is doesn't account for vertical reading directions such as in Japanese newspapers. You would need to draw a wide flat caret instead of the tall thin one here when the DWRITE_READING_DIRECTION was either top-to-bottom or bottom-to-top.
IDWriteTextLayout::GetMetrics only gives the overall bounding box to you, not the caret position.
IDWriteTextLayout::HitTestPoint's isInside flag is true if it is over the text itself, not just the layout bounds.
You can get the layout's bounding rectangle via IDWriteTextLayout::GetMetrics.
DWRITE_TEXT_METRICS textMetrics;
textLayout->GetMetrics(&textMetrics);
Your rectangle is
D2D1::RectF( textMetrics.left,
textMetrics.top,
textMetrics.left + textMetrics.width,
textMetrics.top + textMetrics.height );
You can then draw the caret along the right boundary line.

Magick++ trimming pixels that are not of backgroundColor (pixels same color on top and bottom of image)

The following image of size 1x9 is being trimmed to 1x6 because presumably the pixel at the top is the same color as the pixel at the bottom and in the trim function, these pixels are being identified as the background color, even though the backgroundColor being reported before the execution of the trim function is #FFFFFF.
http://s1.postimage.org/a7r69yxsr/m_medium_bc.png
The only thing I am doing is executing trim on the Image. Explicitly setting backgroundColor and/or transparent() makes no difference.
Why is this occurring and is this the expected behavior?
Can this be fixed by configuration/property setting/without changing Graphicsk library code?
If not, when can this bug be fixed? Do you expect a bug of this nature to be fixed in the next few days?
Here is the code:
Magick::Image tempImage;
tempImage.read(name);
std::cout<<"size:"<<tempImage.columns()<<","<<tempImage.rows()<<std::endl;
temp=tempImage.backgroundColor();
std::cout<<"bg:"<<(std::string)temp<<std::endl;
tempImage.trim();
std::cout<<"size:"<<tempImage.columns()<<","<<tempImage.rows()<<std::endl;
I agree that this behaviour is strange, I am not a developer/maintainer of ImageMagick/Magick++ so can't comment further as to whether this is a bug or 'feature'. However I had the same issue and created this function as a workaround (note this is much faster than manually iterating the pixels, even with a pixel cache in place):
Magick::Geometry CalculateImageMagickBoundingBox( const Magick::Image & image, const Magick::Color & borderColor )
{
// Clone input image.
Magick::Image clone( image );
// Remember original image size.
const Magick::Geometry originalSize( image.columns( ), image.rows( ) );
// Extend geometry by two in width and height (one pixel border).
Magick::Geometry extendedSize( originalSize.width( ) + 2, originalSize.height( ) + 2 );
// Extend cloned canvas (center gravity so 1 pixel border of user specified colour).
clone.extent( extendedSize, borderColor, Magick::CenterGravity );
// Calculate bounding box (will use border colour, which we have set above).
Magick::Geometry boundingBox = clone.boundingBox( );
// We added 1 pixel border, so subtract this now.
boundingBox.xOff( boundingBox.xOff( ) - 1 );
boundingBox.yOff( boundingBox.yOff( ) - 1 );
// Clamp (required for cases where entire image is border colour, and therefore the right/top borders
// that we added are taken into account).
boundingBox.width( std::min( boundingBox.width( ), originalSize.width( ) ) );
boundingBox.height( std::min( boundingBox.height( ), originalSize.height( ) ) );
// Return bounding box.
return boundingBox;
}
In your particular case, you could use this function and then set the canvas size based on the geometry returned.

C++/CLI Visual C++ 2010 Express - Drawing multiple ellipses

I want to draw multiple filled ellipses on/in some panel. Drawing single one isnt problem, i am using:
Color aColor = Color::FromArgb( 255, 0, 0 );
SolidBrush^ aBrush = gcnew SolidBrush(aColor);
Rectangle rect = Rectangle(x, y, 10, 10);
e->Graphics->FillEllipse(aBrush, rect);
It draws red ellipse bordered by rectangle, and fills it with red color. (assuming i will give x and y). The problem i met, is when I want to draw multiple ellipses like that, in RANDOM places. So i need to pass random x and y (using rand() % somenumber) but i am not sure, how can i pass these variables into the panel1_paint function and draw them when both numbers are randomized. Also, ofc i dont want the last ellipse to disappear when drawing new one. The only way is using global variables?
Any ideas?
Well, i tried as suggested, to use loop inside panel and i got that:
for(int i=0; i<ile_przeszkod; i++){
int x = rand() % 690; int y = rand() % 690;
Color aColor = Color::FromArgb( 255, 0, 0 );
SolidBrush^ aBrush = gcnew SolidBrush(aColor);
Rectangle rect = Rectangle(x, y, 10, 10);
e->Graphics->FillEllipse(aBrush, rect);
MessageBox::Show("x: "+x+ " y: " +y);
}
ile_przeszkod means how many of them i want to be drawn, and message box showes me what numbers it randomized so i am sure ellipses dont overlap. The problem is, after "invalidating" panel1 i see only 1 ellipse. :/ What should i do to see both of them?
all the x, y coordinates are random , so they don't depend on some other deciding procedure, So that need not to be passed to panel1_paint rather you can run a lpop and generate random number to use them as your x, y coordinates.

Qt round rectangle, why corners are different?

I try to draw a round rectangle with drawRoundedRect method directly in a QPixmap (no render engine involve here exept pure Qt one ...), I double check the size of the rectangle versus the size of my pixmap :
Pixmap : QSize(50, 73)
Rectangle: QRect(0,0 48x11)
See plenty of space ...
EDIT: some code
pixmap = QPixmap(50,73); //example size that match my case
QRectF rect(0,0,48,11);
QPainter painter(&pixmap);
painter.setRenderHint(QPainter::TextAntialiasing);
painter.setWorldMatrixEnabled(false);
painter.setPen(QPen()); //no pen
painter.setBrush(QBrush(color));
painter.drawRoundedRect(rect, 2.0, 2.0);
I disabled world transformation ...
I set set transformation to unity ...
I tried several radius (1.0,2.0,3.0,4.0) ...
I change pen width, brush color ...
But it always ends with a rectamgle with 4 diferent corners ! Like that :
I directly ouptut the pixmap to a file to be sure I wasn't scraping it during the display ... same shape.
Anyone know about Qt round rectangle with small radius ? I saw somthing about it a long time ago but I don't remenber how to deal with it !
It looks like you're not using anti-aliasing (i.e. the QPainter::Antialiasing render hint). This is a Qt quirk that occurs without it. From what I've seen/heard, the Qt devs aren't terribly concerned with fixing this (most people want anti-aliasing anyway).
The work-around (besides just using anti-aliasing) is to draw the rect yourself with QPainter::drawLine() and QPainter::drawArc(). You might have to play with numbers until it looks right -- straight calculations tend to come out a pixel or two off. Also, you might find that even with this method the lower right corner is never exactly the same as the other corners.
If you're feeling mildly ambitious, you could try fixing this and submitting a patch to Qt.
Update: Arc drawing results changed in Qt 5. In my experience, it's a big improvement.
I know this is an old problem but for Qt5 users calling setRenderHint(QPainter::Qt4CompatiblePainting); on the QPainter seems to solve the problem.
Edit:
I found a solution for generating a perfect rounded rectangle together with border color and it looks the same as the rounded rectangles used by QPushButton's border for example. This is how I implemented the paintEvent to achieve this:
void MyButtonGroup::paintEvent(QPaintEvent * e)
{
int borderSize = 5;
QColor borderColor = Qt::red;
QColor backgroundColor = Qt::blue;
int borderRadius = 3;
QPen pen;
pen.setWidth(borderSize);
pen.setColor(borderColor);
QPainter painter(this);
painter.setRenderHint(QPainter::Antialiasing);
painter.setPen(pen);
QRectF rect(rect().x() + borderSize / 2,
rect().y() + borderSize / 2,
rect().width() - borderSize,
rect().height() - borderSize);
if(borderSize % 2 == 0)
{
painter.drawRoundedRect(rect,
borderSize,
borderSize);
}
else
{
painter.drawRoundedRect(rect.translated(0.5, 0.5),
borderRadius,
borderRadius);
}
QBrush brush(backgroundColor);
pen.setBrush(brush);
painter.setBrush(brush);
if(borderSize % 2 == 0)
{
painter.drawRoundedRect(rect,
borderRadius,
borderRadius);
}
else
{
painter.drawRoundedRect(rect.translated(0.5, 0.5),
borderRadius,
borderRadius);
}
QWidget::paintEvent(e);
}
I'm posting this because I found it a bit hard to achieve this result:
Try adding half a pixel offset (e.g.: rect.translated(0.5,0.5) ):
QRectF rect(0,0,48,11);
painter.setRenderHint(QPainter::Antialiasing,false);
painter.drawRoundedRect( rect.translated(0.5,0.5), 2.0, 2.0 );
I suppose this has to do with the coordinate system placing an integer value between two pixels.
If you draw with antialiasing and use a pen of 1 pixel width then drawing at exact integer coordinates results in lines of 2 pixel width instead.
Only with this 0.5 pixel offset you'll get lines that are exactly 1 pixel wide.
QRectF rect(0,0,48,11);
painter.setRenderHint(QPainter::Antialiasing,true);
painter.setBrush(Qt::NoBrush);
painter.setPen( Qt::white );
painter.drawRoundedRect( rect.translated(0.5,0.5), 2.0,2.0 );
Best way do draw RoundRect is Path.
http://developer.nokia.com/community/wiki/Qt_rounded_rect_widget
void fillRoundRect(QPainter& painter, QRect r, int radius)
{
painter.setRenderHint(QPainter::Antialiasing,true);
QPainterPath rounded_rect;
rounded_rect.addRoundRect(r, radius, radius);
painter.setClipPath(rounded_rect);
painter.fillPath(rounded_rect,painter.brush());
painter.drawPath(rounded_rect);
}
try to play with render hints
1) disable antiAliasing;
2) enable SmoothPixmapTransform
but still no guarantee that it will help.
I have tried all tips from answers here but nothing works for me. But based on these code snippets I have found following solution:
As default set m_pPainter->setRenderHint(QPainter::Qt4CompatiblePainting, true) and only for rounded rectangles with width%2==0 disable it.
QRect rect = ConvertRectangle(rectangle);
int nPenWidth = m_pPainter->pen().width();
if ( nPenWidth % 2 == 0 )
m_pPainter->setRenderHint(QPainter::Qt4CompatiblePainting, false);
m_pPainter->drawRoundedRect(rect, dbRadiusX, dbRadiusY);
if ( nPenWidth % 2 == 0 )
m_pPainter->setRenderHint(QPainter::Qt4CompatiblePainting, true);