(29/11/2010 04:47 PM)ZiNgA BuRgA Wrote:  

(29/11/2010 06:56 AM)Assassinator Wrote:  Downscaling simplified is just like taking the average

Hardly.  Perhaps only the case with something like a bilinear scale.  Certainly not the case with lanczos.

Ok, maybe "average" is the wrong word for that.  What I mean is whatever method, you're doing some type of interpolation, which is akin to fitting some sort of curve or system of curves through a set of points.

In the extreme case, with a dumb nearest neighbor method, you're just dropping an evenly spread percentage of all points (eg. ever second point for a 2x downsize).  That method produces the sharpest output, and the resulting data still contain exactly the same percentage of artefacting relative to the original, since the point dropping method does not discriminate between "good" or "bad" points.

Any other interpolator, which produces a curve smoother than a staircase (ie. everything else), should be less effected by "special" points than the nearest neighbor method, thus the artefacting would get "smoothed away" to some degree compared to the original.

(29/11/2010 04:47 PM)ZiNgA BuRgA Wrote:  

(29/11/2010 06:56 AM)Assassinator Wrote:  EDIT: Plus, JPG doesn't cause haloing, so the problems don't match anyway.

It doesn't, but the combination of everything can result in it.

Even if it does, the part that it played should be so small it pretty much wouldn't matter.  Sharp downsizing always results in haloing, comes with the territory, however lossless the source.

(This post was last modified: 29/11/2010 07:12 PM by Assassinator.)