Density

Threshold the gradients to select three kinds of field

A particle with smallest gradients uses arbitrary field; the middle group uses smoother field; the particles in highest group use the real field.Hard thresholds cannot make sure the good segmentation.

I am working on the calculation of textureness.

Based on the values of gradients to choose different fields

The particle at lower gradients(T=20) chooses the smoother field.

Stroke-based NPR based on current results

There are some wrong trails crossing through edges, which should be deleted or partitioned later.http://www.mrl.nyu.edu/projects/npr/painterly/

Display different levels of lines

Black lines represent the first level. Original lines are for the second level. Green lines are the extensions from previous two levels.

Merge the smoothest version

The first level is using the gently smoothed  field to make it closer to real field. But the second level is using the smoothest field.

We can see some influences but the quality of the trails should be improved.

The smoothest version

Almost-connected network

Fixed up most of unconnected lines. If you found one, let me know it.

Some things:

1.Not happy with the strange shapes I mentioned before.

2.Lower densities of display is not good for structures.

3.Effects are not as good as previous blog. One reason might be that current results are not following the tone. Another might come from the construction of the network since in order to attain a connected network, there are some modifications to maintain the program to stop properly.

   a. Set zero mags to a very small value

   b. Set zero direction to a arbitrary direction.(This one seems rare to happen.)

   c. Try not to allow self-crossed trail, which shoots the particle at intersection with current state again. This one fixed up a lot situations.

   d. For very short trails, clean it up and shoot the particle again with higher speed.Fixed up a lot short lines.

Show magnitude of ETF field

It turns out the direction is more useful than mag. Maybe the current way without considering the mag is ok for our purpose. 

Red: mag = 0 and x direction and y direction = 0

Blue: mag = 0

Green: x direction and y direction = 0(Rarely happen after smoothing)

Greyscale: The strength of ETF field