Filtering Solid Gabor Noise
Supplemental Material

Fig. 1.

For the videos below, we use the FFV1 codec, a lossless video codec.
This is because lossy video codecs can remove subtle aliasing artifacts and can interfere with aliasing patterns.
We suggest to watch these videos using Windows Media Player after installing ffdshow (Files -> Official releases -> generic build) (Windows), using VideoLAN's VLC media player (Windows, Linux, Mac OS X), or using MPlayer (Windows, Linux, Mac OS X).

Sharp Edges

Below we show the videos corresponding to Fig.1(a.1), Fig.1(b.1), Fig.1(c.1) and Fig.1(d.1).
These videos show the noise in the presence of sharp edges.
For Gabor noise and our noise we show both the isotropic as well as anisotropic versions.
These videos show that only our noise and Perlin noise preserve continuity over sharp edges.

Perlin	Wavelet	Gabor	Ours
fig1_a1.avi	fig1_b1.avi	fig1_c1_iso.avi	fig1_d1_iso.avi
		fig1_c1.avi	fig1_d1.avi

Filtering

Below we show the videos corresponding to Fig.1(a.2), Fig.1(b.2), Fig.1(c.2) and Fig.1(d.2).
These videos show a side-by-side comparison of the ground-truth filtered and the filtered noise.
For Gabor noise and our noise we show both the isotropic as well as anisotropic versions.
These videos show that only our noise and Gabor noise support high-quality anisotropic filtering.

The ground-truth filtered noise was obtained using 256x stratified super-sampling.
Note that, despite this large sampling rate, the ground-truth filtered noise still exhibits aliasing in some cases (e.g., in the anisotropic version of Gabor noise and our noise).
However, we were not able to further increase the sampling rate due to the limited precision of the accumulation buffer.
Also note that the ground-truth filtered noise is roughly 256x slower than the filtered noise.

Perlin	Wavelet	Gabor	Ours
fig1_a2.avi	fig1_b2.avi	fig1_c2_iso.avi	fig1_d2_iso.avi
nbsp;	nbsp;	fig1_c2.avi	fig1_d2.avi

Below we show a variant of the videos corresponding to Fig.1(a.2), Fig.1(b.2), Fig.1(c.2) and Fig.1(d.2).
These videos show a side-by-side comparison of the unfiltered and the filtered noise.
For Gabor noise and our noise we show both the isotropic as well as anisotropic versions.

Perlin	Wavelet	Gabor	Ours
fig1_a2.avi	fig1_b2.avi	fig1_c2_iso.avi	fig1_d2_iso.avi
		fig1_c2.avi	fig1_d2.avi

Below we show a comparison based on the videos above.
These videos show a side-by-side comparison of the filtered noise for Perlin noise, wavelet noise and Gabor noise versus our noise.
For Gabor noise versus our noise we show both the isotropic as well as anisotropic versions.

Perlin vs Ours	Wavelet vs Ours	Gabor vs Ours
fig1_a2_vs_d2_iso.avi	fig1_b2_vs_d2_iso.avi	fig1_c2_iso_vs_d2_iso.avi
		fig1_c2_vs_d2.avi

Fig. 6.

Please note that we have modeled the Parthenon scene to scale, and that we have chosen viewpoints that are plausible for a human observer.
(We have however moved both viewpoints up, to make Fig.6 more clear.)

Sharp Edges

Below we show the images corresponding to Fig.6(c.1), Fig.6(c.2), Fig.6(c.3) and Fig.6(c.4).
These images show the texture in the presence of sharp edges.
These images show that only our noise and Perlin noise preserve continuity across sharp edges.

Lighting was disabled to avoid the introduction of discontinuities due to lighting.
We also show the same image without the texture and with lighting.

Perlin	Wavelet	Gabor	Ours
fig6_c1.png	fig6_c2.png	fig6_c3.png	fig6_c4.png

fig6_c_geom.png

Below we show a comparison based on the images above.
These images show a back-and-forth comparison of the texture in the presence of sharp corners for Perlin noise, wavelet noise and Gabor noise versus our noise.

For the back-and-forth comparisons, we use the GIF format, since it supports back-and-forth animation.
Note that the GIF format is limited to 256 colors. Color quantization and dithering can introduce artifacts. Refer to the images above for artifact-free images.

Perlin vs Ours	Wavelet vs Ours	Gabor vs Ours
fig6_c1_vs_c4.gif	fig6_c2_vs_c4.gif	fig6_c3_vs_c4.gif

Filtering

Below we show the images corresponding to Fig.6(d.1), Fig.6(d.2), Fig.6(d.3) and Fig.6(d.4).
These images show the filtered texture.
These images show that only our noise and Gabor noise support high-quality anisotropic filtering.

Lighting was disabled in order to avoid the introduction of gradients due to lighting.
We also show the same image without the texture and with lighting.

Perlin	Wavelet	Gabor	Ours
fig6_d1.png	fig6_d2.png	fig6_d3.png	fig6_d4.png

fig6_d_geom.png

Below we show a comparison based on the images above.
These images show a back-and-forth comparison of the filtered texture for Perlin noise, wavelet noise and Gabor noise versus our noise.

Perlin vs Ours	Wavelet vs Ours	Gabor vs Ours
fig6_d1_vs_d4.gif	fig6_d2_vs_d4.gif	fig6_d3_vs_d4.gif

Fig. 6 bis.

Fig. 6 bis is another version of Fig. 6, which uses a more controlled setup.

Sharp Edges

The camera is looking directly at a horizontal sharp edge.

Perlin	Wavelet	Gabor	Ours
fig6bis_c1.png	fig6bis_c2.png	fig6bis_c3.png	fig6bis_c4.png

Perlin vs Ours	Wavelet vs Ours	Gabor vs Ours
fig6bis_c1_vs_c4.gif	fig6bis_c2_vs_c4.gif	fig6bis_c3_vs_c4.gif

Filtering

The camera is looking at the "horizon" of an "infinite" plane.

Perlin	Wavelet	Gabor	Ours
fig6bis_d1.png	fig6bis_d2.png	fig6bis_d3.png	fig6bis_d4.png

Perlin vs Ours	Wavelet vs Ours	Gabor vs Ours
fig6bis_d1_vs_d4.gif	fig6bis_d2_vs_d4.gif	fig6bis_d3_vs_d4.gif