TMaskCleanerMod for VapourSynth

A really simple mask cleaning plugin for VapourSynth based on mt_hysteresis, also known as a Connected Component Labeling filter.
It discards all areas of less than length pixels with values bigger or equal to thresh. Original plugin by tp7. Ported to VapourSynth by Beatrice-Raws and modified by dtlnor.

Requirements

• VapourSynth R54+ (API 4)
• System: 64bit
• CPU: AVX2+

Example Usage

TMaskCleanerMod

core.tmcm.TMaskCleanerMod(clip clip, [int length, int thresh, int fade, bint binarize, int connectivity, bint reverse, int mode])

tmcm.TMaskCleanerMod(clip, length=5, thresh=235, fade=0)

Original plugin works only with 8 bit input (and only Y plane). Beatrice-Raws port should support 10-16 bit as well, my mod adds 32bit float support on top of it (only plane 0 will be processed, 1 and 2 will be copied).

This mod adds options inspired by OpenCV's CCL based on the Beatrice-Raws port.
See the Syntax and Parameters section for details.

GetCCLStats

core.tmcm.GetCCLStats(clip clip, [int thresh, int connectivity])

tmcm.GetCCLStats(clip, thresh=235)

Get the following FrameProps:

_CCLStatNumLabels
_CCLStatAreas
_CCLStatLefts
_CCLStatTops
_CCLStatWidths
_CCLStatHeights
_CCLStatCentroids_x
_CCLStatCentroids_y

These contain arrays (except _CCLStatNumLabels) if the input clip isn't blank. Note that the first label is always the background label, so you'll typically access frame properties like this:

f.props.get('_CCLStatAreas', None)[1:]

Syntax and Parameters

• clip
  Input clip. Supports 8-16 bit integer / 32bit float Gray/YUV formats. For YUV input, only plane 0 is processed; planes 1 and 2 are copied.

• length = 5
  Discards connected areas that less than length pixels. (Use reverse to discard areas larger than length)

• thresh = 235
  Binarize threshold. Discards pixels with values less than thresh before applying the length filter. Range: 0-255 for 8-bit input, 0-65535 for 16-bit input etc. Default 235 << (bitsPerSample - 8) for 8-16bit or 1.0 for 32bit float input.

• fade = 0
  Assume input clip is 8bit.
  Controls gradual transition from 0 to 255:

  • Areas below length will be zeroed
  • Areas within [length, length+fade] will be assigned values gradually from 0 to 255
  • Areas above length+fade will be set to 255

  This allows adaptive processing of area ranges.
  For example, transforming the mask from linear (0→255) distribution denoted as / to:

  • /\ distribution (0→128→0)
    using .Expr("x 128 > 255 x - x ?")
    = if x > 128 set 255-x, else set x
  • || to grab some arbitrary range
    using .Expr("x 118 > x 138 < & 255 0 ?")
    = only if x > 118 && x < 138 set 255, else set 0
  • Non-linear _/_ mask by smoothly amplifying the middle
    using .Expr("x 128 > 255 x - x ? 100 - 28 / sqrt 256 *")
    = sqrt(/\ - 100 / 28) * 255

• binarize = false

  • false: Retain original luma values before applying length filter and fade multiplication.
    This means the output of retained areas will have their values copied from the source clip.
  • true: Apply binarization with thresh before applying length filter and fade multiplication.
    This means the output of retained areas will be changed to peak value (255 for 8-bit, 65535 for 16-bit, 1.0 for 32-bit float)

• connectivity = 8
  Determines how connected components are identified:

  • 8: Include all 8 surrounding pixels (including diagonals) in a 3×3 neighborhood
  • 4: Include only adjacent pixels (top/bottom/left/right) in a 3×3 neighborhood

• reverse = false
  Set to true to filter out objects larger than length.
  Assume input clip is 8bit.

  • false:
    • Values below length will be zeroed
    • Values within [length, length+fade] will be assigned values gradually from 0 to 255
    • Values above length+fade will be set to 255
  • true:
    • Values above length will be zeroed
    • Values within [length-fade, length] will be assigned values gradually from 255 to 0
    • Values below length-fade will be set to 255

• mode = 0
  Determines how connected components are filtered.
  Note that the meaning of length changes based on the mode.

  • 0: Filter by area of component
  • 1: Filter by centroid x position
  • 2: Filter by centroid y position
  • 3: Filter by left position of bounding box
  • 4: Filter by top position of bounding box
  • 5: Filter by right position of bounding box
  • 6: Filter by bottom position of bounding box
  • 7: Filter by width of bounding box
  • 8: Filter by height of bounding box

License

This plugin is licensed under the MIT license. Binaries are GPL v2 because if I understand licensing stuff right (please tell me if I don't) they must be.