Curtis L. Olson wrote: > This can get to be a really complicated subject if we let it. But, > as it works now, a single full res texture is stored on the disk. > It is loaded at run time, and the mipmaps are generated using a > simple 2x2 -> 1x1 averaging scheme. (I think this is called a box > filter.)
Yeah. This is considered the "naive" way to do it because it's easy to get aliasing artifacts if notable features in the input image are aligned on power-of-two boundaries. I know that Steve Baker likes the idea of FFTing the data first and chopping off the high frequencies in fourier coordinates. This *also* has aliasing artifacts, but they look different. Sharp edges in the input data get "ringing" structures around them, not unlike what you see with JPEG decompression. I've heard of "jittered" averaging too, which tries to avoid aliasing artifacts statistically by picking slightly random offsets during the averaging (this is sort of the opposite of the jittered supersampling that gets done with CG rendering). Needless to say, which algorithm you choose is going to depend in very sensitive ways on your texture image. And then, of course, your choice of LOD bias and texture anisotropy needs to be factored in. My $0.02, obviously, is to forget the whole issue and spend your time on image quality issues that matter, like fancier Airplane models or the distracting pixelation on the current A-4 textures. :) Andy _______________________________________________ Flightgear-devel mailing list [EMAIL PROTECTED] http://mail.flightgear.org/mailman/listinfo/flightgear-devel
