On Aug 15 11:44, Charles Wilson wrote: > On 8/15/2011 10:33 AM, Warren Young wrote: > > The hard edges in the original art are causing stair-stepping when doing > > a direct downsample, though. (Look at the pointy bits.) By blurring > > the high-res version and then downsampling by a non-integral amount, you > > can get a much smoother result. > > This is only the case if the "downsample" operation used by GIMP, when > d/s by an integral amount, is to simply pick every Nth pixel. That's > very fast -- but is not the correct operation (I'd posit a GIMP bug, in > fact). > > Sampling theory says a downsample SHOULD be preceded, automatically, by > a low-pass filter (blurring) operation of a specific type and, er, > "radius" for lack of a better word. (IOW, GIMP /should/ be doing this > blur FOR you, automatically). There's lots of theory behind this, to > select the proper kind of filter (gaussian is not correct -- but is > probably a good enough approximation) and its 'radius' (which should > scale with the downsampling factor). > > Since GIMP is apparently not doing that, then yes -- you need to apply a > blurring filter yourself, before using GIMP's braindead 'pick every Nth > pixel' version of "downsampling".
Are you talking about recent gimp versions? In my gimp I have the choice of four different interpolation algorithms, "None", "Linear", "Cubic", and "Sinc (Lanczos3)", whatever each of them means. I guess I just don't want to know in such great detail... Corinna -- Corinna Vinschen Please, send mails regarding Cygwin to Cygwin Project Co-Leader cygwin AT cygwin DOT com Red Hat