Tuukka wrote:
> Maybe you could convert the images in-place: that is, the conversion
> routine wouldn't copy the image and thus the green pixels wouldn't need
> touched at all (this should already give 30% speedup). Furthermore, cache
> should work better since you would need only half the memory.
I tried that, but I get segmentation faults. This is probably because by the
time the buffer has been converted, it is also accessed by the frame grabber
capturing a new frame. I tried to put it within a mutex block, but that
didn't work either.
The function I used for that is (courtesy of videodog 0.21):
void bgr24_to_rgb24(unsigned char *buf, int size)
{
unsigned char c;
unsigned char *p = mem;
int i = size;
while (--i)
{
c = p[0];
p[0] = p[2];
p[2] = c;
p += 3;
}
}
/G
