|
sepele
Figure 3 This method does a marvelous job of synthesizing the sepele bark image. Though it obviously has features with different characteristic size, the image is roughly stationary upon inspection, and its spatial features are not sharply localized in space, and is thus quite easy to synthesize. |  |
|
wood
Figure 4 Excellent performance. Both the texture of the wood and the lineations along its length are synthesized seamlessly. |  |
|
herringbone
Figure 5 This image is simply parameterized in terms of two-point statistics, which our algorithm uses, hence the excellent synthesis of the herringbone fabric image. This result effectively maintains the vertical homogeneity property of the TI. |  |
|
rand-squares
Figure 6 The result is quite good. Many (not all) of the solid black areas in the output look like squares, and the relative density of squares is quite similar. |  |
|
fabric
Figure 7 At a first glance, one might expect a similarly striking result as the herringbone fabric synthesis. Unfortunately, the method was nowhere near as successful, though we don't characterize this result as a failure, either. The differences between this image and the herringbone are subtle: lower contrast in most regions, less regularity within patches of constant weave, some residual curvature in the lower portion of the image. |  |
|
brick
Figure 8 Our method performs surprisingly poorly on this image. The complexity of the localization is severe; for instance, the randomly placed patches of constant gray are localized within individual bricks, and must be so in the SI. |  |
|
polygons
Figure 9 The simplistic appearence of this image is misleading, for it is highly difficult to synthesize. The features which we must synthesize are the edges, which occur at a multitude of different orientations. |  |