Quantcast Nomosmooth Filter for Avisynth [Readme] - digitalFAQ.com Forums [Archives]
  #1  
11-25-2012, 07:00 PM
kvcd.net Archives's Avatar
kvcd.net Archives kvcd.net Archives is offline
Free Member
 
Join Date: May 2012
Posts: 84
Thanks: 0
Thanked 1 Time in 1 Post
NoMoSmooth - An Avisynth filter for motion-adaptive spatio-temporal smoothing

By Ross Thomas <ross@grinfinity.com>

When it comes to smoothers (as opposed to smoothies) there are two main varieties: temporal smoothers and spatial smoothers. Temporal smoothers work by looking at frames surrounding the current one and averaging corresponding pixels together if they're deemed similar enough (and thus likely to be noise). Spatial smoothers work in a similar way, except they look at nearby pixels within the same frame. A few filters are a hybrid of the two, including this one and the excellent Convolution3D.

Each approach has its advantages and disadvantages. The great thing about temporal smoothers is that they do a fantastic job getting rid of noise when you set them up just right, but they can also generate some very ugly artifacts, particularly when dealing with motion. Spatial smoothers, on the other hand, can be quite destructive to the details in your image, but are great with high-motion areas because those have few details and are moving too quickly to notice artifacts anyway.

Ideally, then, one would like to apply a temporal smoother to relatively static areas and a spatial smoother to moving ones. That's what this filter tries to do, and the aim is for a higher quality result (i.e. less noticible smoothing) than Convolution3D but with correspondingly less of an impact on compressibility.

In addition to this motion-based approach, NoMoSmooth employs another technique to try to retain as much existing detail as possible: only pixels that are "fluctuating" are smoothed. In other words, given three frames in a row with the current one in the centre, the pixel being examined will only be considered for either type of smoothing if it is either greater than both the corresponding pixels in the previous and next frames, or less than both. This could be considered a very simplistic "noise detector" and seems to help reduce temporal smoothing on important details such as those in skin tones.

Usage
Code:
NoMoSmooth([clip], int motion_threshold, int temporal_radius, int temporal_threshold, int spatial_radius, int spatial_threshold, bool show)
ParameterMeaningDefault
motion_thresholdControls how sensitive the motion detector is to, er, motion, with lower values being more sensitive (seeing more motion) and higher being less sensitive. I might go into the algorithm in more detail in a later version of this file, but for now if you're curious look at the source.40
temporal_radiusDetermines how far the temporal smoother will venture into the clip to do its work. If set to 2 a total of 5 frames are examined, 2 on each "side" of the current one.1
temporal_thresholdWhen the temporal smoother is averaging it will only include values within this threshold of the pixel in the current frame.6
spatial_radiusDetermines how many pixels the spatial smoother will consider when doing it's job. If set to 2 a total of 25 pixels are examined, with the current pixel in the centre.1
spatial_thresholdThe spatial smoother will only include values within this threshold when it is averaging.3
showIf true no smoothing will be carried out, but areas deemed to be in motion (according to the motion_threshold parameter) will be highlighted in a charming light grey.False

Known Issues
- Pixels within spatial_radius of the edges get passed through.
- Frames within temporal_radius of the ends of the clip get passed through.
- While the algorithm is fairly well optimized it's still fairly slow, but very brief tests indicate it might be slightly faster than Convolution3D in preset="movieHQ" mode.

To-Do
- Optimize both algorithmically and by implementing assembly code versions of critical parts of the algorithm.
- Improve motion detector such that it is more sensitive to motion in low-contrast areas.

History
  • 0.1a - Fixed off-by-one error that caused an access violation when height > 462.
  • 0.1 - Total rewrite with a number of algorithmic improvements, much more verbose comments and widespread use of assertions. Made brief preparations for an eventual YV12 version. Implemented "noise detector". First official (yet still alpha) release.
  • 0.0c - Back on track. Rewrote again, this time making a conscious effort not to duplicate code already in the core. Implemented a motion detector so that areas in motion get spatially softened and static areas get softened temporally. The name "NoMoSmooth" makes less sense every time I read it.
  • 0.0b - Highly embarrassing version algorithmically identical to TemporalSoften which must never be spoken of again.
  • 0.0a - Rewrote from scratch and optimized the algorithm somewhat, so now is slightly faster. A change in the algorithm both removed the need for noise_threshold and made show mode extremely difficult to implement.
  • 0.0 - Proof of concept code. Hard hats must be worn in this area.

Download here: SansGrip's Avisynth Filters [DOWNLOADS]
Reply With Quote
Someday, 12:01 PM
admin's Avatar
Site Staff / Ad Manager
 
Join Date: Dec 2002
Posts: 42
Thanks: ∞
Thanked 42 Times in 42 Posts
Reply




Similar Threads
Thread Thread Starter Forum Replies Last Post
LegalClip Filter for Avisynth [Readme] kvcd.net Archives Avisynth Scripting 0 11-25-2012 09:03 PM
Ghostbuster Filter for Avisynth [Readme] kvcd.net Archives Avisynth Scripting 0 11-25-2012 09:00 PM
Blockbuster Filter for Avisynth [Readme] kvcd.net Archives Avisynth Scripting 0 11-25-2012 07:38 PM
Sampler filter for Avisynth [Readme] kvcd.net Archives Avisynth Scripting 0 11-25-2012 06:24 PM
RemoveBorders filter for Avisynth [Readme] kvcd.net Archives Avisynth Scripting 0 10-28-2012 08:04 AM

Thread Tools



 
All times are GMT -5. The time now is 06:54 PM  —  vBulletin © Jelsoft Enterprises Ltd