Tuesday, April 19, 2011

Results with Real Textures

Here are some results with real textures. I used window radius of 32 (so 64x64 window), which is quite big.

Few problems that will have to be tackled:
  1. Visible square edges around neighborhood window. (See result #1). I thought blending would take care of this but it is showing up quite prominently in the green pattern.
  2. Blended regions are unclear and have less contrast. I will have to look more carefully into how the paper handles overlap regions. They talk bit about accumulation of error in the overlapping region.
Here are two results:


Monday, April 18, 2011

Varying Neighborhood Sizes

Okay! Figured out why this is working. I found this sentence in the paper:
"the neighborhood size should be large enough to capture repeating elements of the texture."
 Here are some results of differing window sizes performed on the same 8x8 square grid. I started with window radius (half width) of 16, then keep decreasing the sizes. The further we get from 16 the more blurry the

window radius: 16
window radius: 15

window radius: 14
window radius: 13

window radius: 12