Auto threshold value of monochrome conversion

[M. Bashir]

Hi All,

It's easy to convert any image to monochrome format by using QImage::convertToFormat like this:

pImage.convertToFormat(QImage::Format_Mono, Qt::MonoOnly|Qt::ThresholdDither|Qt::AvoidDither);
but convertToFormat always guessing that the value of conversion's threshold is 255 where in some cases we need to use another threshold's value. I checked out GIMP and I noticed that it uses an automatic method for guessing threshold's value (see the example below) so I'm wondering:

How I can convert images to monochrome format by using auto threshold's value just like GIMP conversion method?


Original
http://img134.imageshack.us/img134/9741/77363363.th.jpg (http://img134.imageshack.us/i/77363363.jpg/)

QImage conversion
http://img145.imageshack.us/img145/5563/43162085.th.jpg (http://img145.imageshack.us/i/43162085.jpg/)

Automatic GIMP conversion
http://img402.imageshack.us/img402/7148/20634127.th.jpg (http://img402.imageshack.us/i/20634127.jpg/)


P.S

I've wrote a stupid function for converting images to monochrome by using a specified threshold's value as following but I need to know how I can automatically specify threshold's value:

QImage getThresholdBW(QImage pImage, int threshold)
{
QImage result = pImage;
result.fill(255);
for(int x=0; x<pImage.width(); x++)
for(int y=0; y<pImage.height(); y++)
result.setPixel(x, y, qGray(pImage.pixel(x, y))>threshold?qRgb(255, 255, 255):qRgb(0, 0, 0));

return result;
}

[wysota]

It either uses 127 as the threshold or if it is smarter, it first calculates an average or median of lightness (which is an equivalent of colors->levels->auto) (aka value in HSV) and uses that. Try doing colors->levels->auto on your image and then convert it to monochrome using 127 as the threshold. If it works you can repeat the same process in your application.

[M. Bashir]

Try doing colors->levels->auto on your image and then convert it to monochrome using 127 as the threshold. If it works
It works.


you can repeat the same process in your application.
I tried to calculate the median but the resultant image isn't correct because the median value is big (about 203)... see my snippet plz:


QImage result = pImage;
QList<int> pixelsValues;
for(int x=0; x<pImage.width();x++)
{
for(int y=0; y<pImage.height();y++)
{
QRgb rgb = pImage.pixel(x,y);
QColor color(rgb);
pixelsValues<<color.lightness();
}
}
qSort(pixelsValues);
int median;
int count = pixelsValues.count();
if(count%2==0)
median = (pixelsValues[count/2]+pixelsValues[(count/2)+1]) / 2;
else
median = pixelsValues[count/2];
threshold = median;
for(int x=0; x<pImage.width(); x++)
for(int y=0; y<pImage.height(); y++)
result.setPixel(x, y, qGray(pImage.pixel(x, y))>threshold?qRgb(255, 255, 255):qRgb(0, 0, 0));
qDebug("threshold = %d", threshold);
return result;

P.S
I'm using Qt 4.6.0 which has lightness function

[M. Bashir]

You can preview the resultant image down here:

http://img80.imageshack.us/img80/6408/reso.th.png (http://img80.imageshack.us/i/reso.png/)

[wysota]

Try with an average. Median will tend to make 50% of your pixels white and the other 50% black. If you have one purple spot in a completely yellow picture that's probably not what you would want. Average should do better. If that doesn't help, there is a third choice: (minimum+maximum)/2 - this is exactly what colors->levels->auto should be doing.

[M. Bashir]

Average should do better
It gave me nearly same result.


there is a third choice: (minimum+maximum)/2 - this is exactly what colors->levels->auto should be doing.
It's working, thanks wysota.


QImage result = pImage;
QList<int> pixelsValues;
for(int x=0; x<pImage.width();x++)
{
for(int y=0; y<pImage.height();y++)
{
QRgb rgb = pImage.pixel(x,y);
QColor color(rgb);
pixelsValues<<color.lightness();
}
}
qSort(pixelsValues);
// int median;
// int count = pixelsValues.count();
// if(count%2==0)
// median = (pixelsValues[count/2]+pixelsValues[(count/2)+1]) / 2;
// else
// median = pixelsValues[count/2];
int median2 = (pixelsValues.first() + pixelsValues.last())/2;
int threshold = median2;
for(int x=0; x<pImage.width(); x++)
for(int y=0; y<pImage.height(); y++)
result.setPixel(x, y, qGray(pImage.pixel(x, y))>threshold?qRgb(255, 255, 255):qRgb(0, 0, 0));
qDebug("threshold = %d", threshold);
// result.save("c:/res.jpg");
return result;