uberzev
 |
Starting with a Checker...
Plugged into a High Pass of 16 
*** Now I'd like to generate a 1/4 scaled version. *** Photoshop shows us how that should look... 
How about FilterForge at 150x150?  Oops, too chubby! 
Ok how about 600x600, scale the checkers by 1/4, and then apply a high pass of 4?  Still too chubby! 
What about a high pass of 16 and then scale the checkers?  Finally Success! But the filter is now achingly slow! 
3.85 secs vs 0.82 when high pass is the final step. There has to be a better solution! 
PS These are simplified examples. The impetus for starting this thread was being unhappy with the control I had over the cracks in my Cracked Earth filter. High Pass Annoyance.ffxml |
|||||
| Posted: March 10, 2011 3:40 am | ||||||
Sphinx.
 |
You're comparing a resampled 150x150 version of the 600x600 image with the same image rendered at 150x150 natively in Filter Forge, right?
I think what is causing this is the real fractional radius values, combined with that extreme contrast setting (it causes the fractional part to be visually the same as a fully covered sample)... not sure though - could be wrong internal rounding, as many of the thumbnails on the site (and in FF) looks awful when bitmap based components are involved.. |
|||||
| Posted: March 10, 2011 4:11 am | ||||||
|
Sphinx.
|
When you resample the 600x600 image you'll notice that the border areas in the cross gets a minor transition gradient (the anti alias).
In FF if you render that cranked up contrast highpass at 150x150, this area will not be there and depending on the real fractional radius you might see either what looks chubby or thinned .. thats my guess.. |
|||||
| Posted: March 10, 2011 4:15 am | ||||||
|
uberzev
|
Yes
Download the Filter attachment to see for yourself. It's kind of confusing otherwise. |
|||||
| Posted: March 10, 2011 4:28 am | ||||||
|
Sphinx.
|
I did check it out - what I meant with anti-alias during resampling is not really related to Filter Forge, its just a "bi product" of downsampling (since areas are summed).
Its because of the internal contrast adjustment in the high pass - it causes the partly covered (by the high pass window defined by a fractional radius) to get maxed out.. the only way to avoid that I think is to not crank it up to 100 (use 99 instead) and filter the result with a step curve (to get the 100% contrast look) High Pass Annoyance - step curve.ffxml |
|||||
| Posted: March 10, 2011 4:50 am | ||||||
|
uberzev
|
That's a clever solution, but the results are still different. Take a look at this better example inspired by my Cracked Earth filter. I've moved to 1/8 to make the difference more apparent. The fast method clearly loses fine details, even if you correct for the thickness. High Pass Annoyance - Cells.ffxml |
|||||
| Posted: March 10, 2011 5:28 am | ||||||
|
uberzev
|
||||||
| Posted: March 10, 2011 5:47 am | ||||||
|
Sphinx.
|
I think thats an unrelated topic. This is related to how the bitmap based components allocate and render the internal bitmap.
I'm not completely aware of how this works, but there is a great difference in scaling before blurring an visa versa. If you scale down after the blur, you force the blur to internally render a much larger bitmap (in non seamless mode). I'd like to know the exact details behind this mechanism too... |
|||||
| Posted: March 10, 2011 7:58 am | ||||||
Join Our Community!
Filter Forge has a thriving, vibrant, knowledgeable user community. Feel free to join us and have fun!
33,711 Registered Users
+18 new in 30 days!
153,533 Posts
+38 new in 30 days!
15,348 Topics
+73 new in year!
Online Users Last minute:
18 unregistered users.
Recent Forum Posts:
- Manson Grudge Border by Jaret by Minko
3 hours ago - Kaleidoscopic Variations Vol. 2 by jhantares by jhantares
September 10, 2025 - Polygonal Images Set by Ramlyn by westdale
September 10, 2025 - FF Killing Resolution by SpaceRay
September 9, 2025 - Geometry Spinner by Rachel Duim by Rachel Duim
September 6, 2025 - Channel Glitch by Martin Naumann by SpaceRay
August 31, 2025 - Circuit Glitch by Martin Naumann by SpaceRay
August 31, 2025 - Groovy Rays by Ramlyn by Ramlyn
August 31, 2025 - Is Ultraforge dead? by dpankros
August 30, 2025 - Component Instances by AussieDecalf
August 27, 2025 - Hyperbolic Tiling Challenge by AussieDecalf
August 27, 2025 - Chaos Fields by SpaceRay
August 27, 2025
