In graphics we created mip-maps with bilateral filtering. Is there any benefit to doing that vs. using gaussian blurring?
motoole2
Bilateral filtering is particularly useful when it comes to preserving sharp edges, whereas gaussian blurring will cause edges to blur. I am, however, surprised to hear that mipmaps use bilateral filtering; because it preserves high-frequency content near edges, subsampling the signal can still lead to aliasing artifacts---at least in theory. Do you happen to have a reference for this? (Could it be that this is being confused with bilinear interpolation?)
leafcat
Oh yeah, that’s my bad, I meant bilinear interpolation.
motoole2
Ah! I see. Bilinear interpolation is typically used when querying the pixel values in the mip-map, both in graphics and in vision. For example, when sampling non-integer pixel coordinates, bilinear interpolation can be used to compute a value based on the nearest pixels. This is done in addition to gaussian filtering, which is used when generating the mip-map.
In graphics we created mip-maps with bilateral filtering. Is there any benefit to doing that vs. using gaussian blurring?
Bilateral filtering is particularly useful when it comes to preserving sharp edges, whereas gaussian blurring will cause edges to blur. I am, however, surprised to hear that mipmaps use bilateral filtering; because it preserves high-frequency content near edges, subsampling the signal can still lead to aliasing artifacts---at least in theory. Do you happen to have a reference for this? (Could it be that this is being confused with bilinear interpolation?)
Oh yeah, that’s my bad, I meant bilinear interpolation.
Ah! I see. Bilinear interpolation is typically used when querying the pixel values in the mip-map, both in graphics and in vision. For example, when sampling non-integer pixel coordinates, bilinear interpolation can be used to compute a value based on the nearest pixels. This is done in addition to gaussian filtering, which is used when generating the mip-map.