A halftoning for paper discussion

Think about what's new in their post?

Original image

                          300 strokes

                         <1700 strokes

<3400 strokes

                           >50000 strokes

Here are some examples from our method.< 12seconds.

Some related pictures

1.http://www.flickr.com/groups/gmic/discuss/72157624425214456/
2.Artistic Sketching with a Painterly Rendering Algorithm
3. Ours????

Curvature values for each trail

To show the curvature for each trail. But it seems there is something wrong for the calculation.

Focus on the quality of each trail

Here are three trails. But there are a lot of points we do not like.

How to improve the results from the old process without branching and grouping?

Not bad results for plants or flowers. How to detect the main structures? Or how to present or to improve the details? What's missing? What's there?

Variations from branching

More results from branching- Still have a lot of issues

First try for branching

seed=10

seed=40

seed =100

seed = 300

Randomly add some starting seeds.

Spark a group of new particles (p1)

Good thing: Sparking new particles can be used for fixing up small details like eyes. But at the sparking position, we can notice the artifacts. Maybe we can select the shapes for the sparking spot. Right now it is a point. Maybe a line or a region can be better for some situations.

I will spend some time on studying the group behaviours.

An older result

Spark another two particles

 Spark another two particles with different direction from previous setting

Spark another four particles

Another example

Blurring from CAH

It turns out it is not working.

Ours:

I1 = sum((1-W) *I2);

Black pixels are those pixels with only one pixel in a mask.

Original Image

Bilateral Filter:

First testing for T-junction

Red lines are hidden. It seems we should define when T-junctions happen. Not always working for every intersection.

Some issues

Right now I am not worried about the pattern quality a lot. But I am worried about the tone. Structure should be further improved by matching region directions with the pattern directions.

Results from manga screening