[KimDaBa] Red-eye removal

Martin Ehmsen martin at ehmsen.org
Mon Jan 10 18:38:29 CET 2005


* Franck COLLINEAU <franck-collineau at wanadoo.fr> [Jan 05. 2005 07:55]:
> I think kimdaba developper team should meet digikam developper team 
> because digikam offers this feature.
> Is it possible ?

I had a look at the digikam code, and their red-eye removal is quiet 
easy, and could probably easily be implemented in Kimdaba (if Jesper 
think is should).
The only thing that stops me from implementing it in Kimdaba is my 
lacking C++ and KDE programming skills.

Here is the actual red-eye removal code from digikam (the file is called 
imageeffect_redeye.cpp):
CODE_BEGIN:
...
uint* data = iface.getSelectedData();
int   w    = iface.selectedWidth();
int   h    = iface.selectedHeight();
...
uint* newData = new uint[w*h];
memcpy(newData, data, w*h*sizeof(unsigned int));

int   r,g,b,a;

struct channel {
    float red_gain;
    float green_gain;
    float blue_gain;
};
channel red_chan, green_chan, blue_chan;

red_chan.red_gain   = 0.1;
red_chan.green_gain = 0.6;
red_chan.blue_gain  = 0.3;

green_chan.red_gain   = 0.0;
green_chan.green_gain = 1.0;
green_chan.blue_gain  = 0.0;

blue_chan.red_gain   = 0.0;
blue_chan.green_gain = 0.0;
blue_chan.blue_gain  = 1.0;

float red_norm, green_norm, blue_norm;

red_norm = 1.0/(red_chan.red_gain + red_chan.green_gain + red_chan.blue_gain);
green_norm = 1.0/(green_chan.red_gain + green_chan.green_gain + green_chan.blue_gain);
blue_norm = 1.0/(blue_chan.red_gain + blue_chan.green_gain + blue_chan.blue_gain);

uint* ptr  = data;
uint* nptr = newData;

int r1, g1, b1;

for (int i=0; i<w*h; i++) {
    a = (*ptr >> 24) & 0xff;
    r = (*ptr >> 16) & 0xff;
    g = (*ptr >> 8)  & 0xff;
    b = (*ptr)       & 0xff;

    if ( r >= ( 2 * g) ) {
        r1 = (int) QMIN(255, red_norm * (red_chan.red_gain   * r +
                                   red_chan.green_gain * g +
                                   red_chan.blue_gain  * b));
        b1 = (int) QMIN(255, green_norm * (green_chan.red_gain   * r +
                                     green_chan.green_gain * g +
                                     green_chan.blue_gain  * b));
        g1 = (int) QMIN(255, blue_norm * (blue_chan.red_gain   * r +
                                    blue_chan.green_gain * g +
                                    blue_chan.blue_gain  * b));
        *nptr = QMIN(int((r-g)/150.0*255.0),255) << 24 | r1 << 16 | g1 << 8 | b1;
    }
    ptr++;
    nptr++;
}
CODE_END.

then newData contains the selected part of the orig. image without the 
red-eye.

So what I would like is:
1. I open an image for viewing and zooms in on a red-eye.
2. I right-click on the image and from the menu chooses "Remove red-eye"
3. The above code is run on the zoomed in portion of the image, and the 
   result is displayed.
4. A dialog-box appears and asks: "Save", "Save to new", "Cancel"
5. The appropriate code is run depending on the users choice.

I think it should be fairly straight forward for anyone knowing Kimdaba 
internals and QT/KDE to implement.

Anyone have any thought about this? Jesper??
Is this code too close to making Kimdaba an image manipulation 
software?!?

Martin

P.S. Sorry Franck.
-- 
The shortest math joke: let epsilon be < 0



More information about the KimDaBa mailing list