QwtRasterData and non-linearly sampled data

[DizzyMoods]

Hello. I want some suggestions on how to my code a bit more versatile. I altered the SpectrogramData example for a discrete series of XYZ data simply by coupling the x and y to the discrete values of my Data->ZData vector.

class SpectrogramData: public QwtRasterData
{
public:
...
    virtual double value(double x, double y) const
    {
        double dx, dy;
        dx = (Data->Xdata.last()-Data->Xdata.first())/(Data->Xdata.size()-1);
        dy = (Data->Ydata.last()-Data->Ydata.first())/(Data->Ydata.size()-1);
        int i = (x-Data->Xdata.first())/dx, j = (y-Data->Ydata.first())/dy;
        return Data->Zdata[j][i];
    }
...
}

class SpectrogramData: public QwtRasterData
{
public:
...
    virtual double value(double x, double y) const
    {
        double dx, dy;
        dx = (Data->Xdata.last()-Data->Xdata.first())/(Data->Xdata.size()-1);
        dy = (Data->Ydata.last()-Data->Ydata.first())/(Data->Ydata.size()-1);
        int i = (x-Data->Xdata.first())/dx, j = (y-Data->Ydata.first())/dy;
        return Data->Zdata[j][i];
    }
...
}

To copy to clipboard, switch view to plain text mode 

This type of resampling works well if I am operating with linearly sampled data (XData and YData are equidistant). How can I improve this algorithm for a series of non-linearly scaled data (i.e. XData = [1 10 20 50 70 100 1000])?

[Uwe]

This type of resampling works well if I am operating with linearly sampled data (XData and YData are equidistant.

One comment on your code: Spectrogram::value is called very often ( for each pixel ! ) and you should try to implement it as fast as possible. Better calculate dx and dy once instead of recalculating it again and again.

How can I improve this algorithm for a series of non-linearly scaled data (i.e. XData = [1 10 20 50 70 100 1000])?

Your code implements a nearest neighbor resampling. You can do the same for XData - beside the algorithm how to identify the next neighbor is different.

Again you should be aware of the fact, that Spectrogram::value is called very often and heavy calculations shouldn't be avoided if possible. F.e. you could overload QwtRasterData::initRaster() and QwtRasterData::discardRaster() to initialize some values, that are constant for the situation - or build a temporary linear value matrix.

Uwe

[DizzyMoods]

Thanks, Uwe.
Now I have some question about the initraster() function. Should I reimplement the whole function (similar to value) or should I just call it with particular QRectF and QSize values? This is my rendition SpectrogramData:

class SpectrogramData: public QwtRasterData
{
public:
    SpectrumData *Data;
    double dx, dy;
    SpectrogramData(SpectrumData *InputData)
    {
        Data = InputData;
        setInterval( Qt::XAxis, QwtInterval( Data->.first(), Data->XData.last() ) );
        setInterval( Qt::YAxis, QwtInterval( Data->YData.first(), Data->YData.last() ) );
        setInterval( Qt::ZAxis, QwtInterval( Data->MinZ, Data->MaxZ ) );
        dx = (Data->XData.last()-Data->XData.first())/(Data->XData.size()-1);
        dy = (Data->YData.last()-Data->YData.first())/(Data->YData.size()-1);
    }
 
    virtual double value(double x, double y) const
    {
        int i = (x-Data->XData.first())/dx, j = (y-Data->YData.first())/dy;
        return Data->ZData[j][i];
    }
}

class SpectrogramData: public QwtRasterData
{
public:
    SpectrumData *Data;
    double dx, dy;
    SpectrogramData(SpectrumData *InputData)
    {
        Data = InputData;
        setInterval( Qt::XAxis, QwtInterval( Data->.first(), Data->XData.last() ) );
        setInterval( Qt::YAxis, QwtInterval( Data->YData.first(), Data->YData.last() ) );
        setInterval( Qt::ZAxis, QwtInterval( Data->MinZ, Data->MaxZ ) );
        dx = (Data->XData.last()-Data->XData.first())/(Data->XData.size()-1);
        dy = (Data->YData.last()-Data->YData.first())/(Data->YData.size()-1);
    }

    virtual double value(double x, double y) const
    {
        int i = (x-Data->XData.first())/dx, j = (y-Data->YData.first())/dy;
        return Data->ZData[j][i];
    }
}

To copy to clipboard, switch view to plain text mode 

I tried to experiment with your suggestion and added these lines to the end of the constructor (lines 14+):

QSize QS(Data->LengthOfX,Data->LengthOfY);
QRectF QR(0,0,500,500); // Some random numbers to initialize QRectF 
initRaster(QR,QS);

QSize QS(Data->LengthOfX,Data->LengthOfY);
QRectF QR(0,0,500,500); // Some random numbers to initialize QRectF 
initRaster(QR,QS);

To copy to clipboard, switch view to plain text mode 

but this implementation doesn't seem to yield any difference. Could you elaborate a bit more on initraster()? I searched this forum, but I can't find a good example of its usage.
Thanks in advance.

[Uwe]

No you don't have to call initRaster - you have to implement it ( like the value method ) !

The raster data object gets initialized by the spectrogram item each time before the image composition gets started. F.e. when you have a huge amount of data ( think of GIS data like the elevation for Canada every 10x10m ) you simply can't have the complete raster in memory. Here an implementation of ElevationData::initRaster would load the values for the requested tile from disk into a 2D matrix.

Your implemention of SpectrogramData::initRaster knows about the raster ( bounding rectangle and the size of the data pixels ) in advance and you could build a matrix ( resample your ~logarithmic matrix ) so, that you can have fast lookups in SpectrogramData::value().

But this all doesn't change the fact that you have to implement the resampling in your code.

Uwe