17 |
} |
} |
18 |
\end{figure*} |
\end{figure*} |
19 |
|
|
20 |
|
|
21 |
|
The most important area for reading is naturally the center of the fisheye, |
22 |
|
where the transformation is nearly orthonormal - the |
23 |
|
edges are mostly used for getting a sense of the context, not for reading. |
24 |
|
|
25 |
|
After some investigation, we discovered that we had found a special case |
26 |
|
of a general principle: if a texture image is only transformed through |
27 |
|
rotation and isotropic scaling, a better filtering result is always obtained |
28 |
|
by applying the stretch-squish operation. |
29 |
|
|
30 |
|
\begin{figure} |
31 |
|
\begin{tabular*}{\columnwidth}{rc} |
32 |
|
a) & \includegraphics[width=5cm]{footprint.1} \\ |
33 |
|
b) & \includegraphics[width=4cm]{footprint.2} |
34 |
|
\end{tabular*} |
35 |
|
\caption{ |
36 |
|
\label{figfootprint} |
37 |
|
Pixel footprint in screen space (PFSS) diagram. |
38 |
|
Texture samples' contribution to a pixel's value. |
39 |
|
a) An explanation of PFSS diagrams: the diagrams |
40 |
|
show the contribution of each texel to the pixel |
41 |
|
as a color (black = no contribution, white = large contribution). |
42 |
|
b) An example PFSS of an EWA texture filterer without |
43 |
|
mipmapping (mockup, just diagrammatic). |
44 |
|
In screen space, the |
45 |
|
filter is circular and has soft edges, while in texture space it would be |
46 |
|
elliptical. |
47 |
|
} |
48 |
|
\end{figure} |
49 |
|
|
50 |
|
Figure~\ref{figfootprint} shows a legend of PFSS diagrams and |
51 |
|
a diagram for the EWA filtering method. |
52 |
|
Appendix A shows how |
53 |
|
PFSS diagrams can be generated easily to show the actual behaviour of a hardware accelerator. |
54 |
|
|
55 |
|
In Section~\ref{seccomp}, we compare the performance of different filtering |
56 |
|
methods, including trilinear, stretch-squish aniso and supersampling, on a test image. |
57 |
|
|
58 |
|
|
59 |
|
- For text, setting of the problem: orthogonal transformations are most important, |
60 |
|
TrueType shows maybe not the right model but ... |
61 |
|
|
62 |
|
|
63 |
|
- case we consider: sharp edges, orthogonal (or nearly so) transformations, e.g. text |
64 |
|
|
65 |
|
- simple solution for improving the situation in one direction: |
66 |
|
stretch the texture in one direction, squish back by texture coordinates. activate |
67 |
|
the aniso filter. Aniso filters planned so that they don't flicker, either. |
68 |
|
|