Changes between Version 2 and Version 3 of denoise


Ignore:
Timestamp:
Jun 9, 2024, 4:23:21 AM (2 years ago)
Author:
MasterQuestionable
Comment:

͏    Public intended.
͏    Revert for apparent broken formatting rendered.

͏    No element is trivially chosen.
͏    Each design has functional purpose.

͏    More related: ​https://github.com/MasterInQuestion/talk/discussions/28

Legend:

Unmodified
Added
Removed
Modified

  • denoise

    v2 v3  
    1 = Denoise
     1͏= Denoise =
    22
    3     Denoise is removing the signal insignificant or unsensible [1] in intended view.
    4     Which shall improve compressibility due to reduced entropy (effective amount of information); and subjectively improve quality.
     3͏    Denoise is removing the signal insignificant or unsensible [1] in intended view.\\
     4͏    Which shall improve compressibility due to reduced entropy (effective amount of information); and subjectively improve quality.\\
     5[ [1]\\
     6͏    "insignificant" is that cannot be seen.\\
     7͏    "unsensible" is that does not make sense. ]
    58
    6 [ [1] "insignificant" is that cannot be seen.  "unsensible" is that does not make sense. ]
     9͏    Typical denoise is value clamp based:\\
     10͏    Transform the data into frequency domain (via Fourier Transform alike), or Wavelet:\\
     11͏    [ https://scikit-image.org/docs/stable/auto_examples/filters/plot_denoise_wavelet.html ]\\
     12͏    And allow passing only certain range.
    713
    8     Typical denoise is value clamp based:
    9     Transform the data into frequency domain (via Fourier Transform alike), or Wavelet:
    10     [ https://scikit-image.org/docs/stable/auto_examples/filters/plot_denoise_wavelet.html ]
    11     And allow passing only certain range.
     14͏    FFT (Fast Fourier Transform) is fast implementation of DFT (Discrete Fourier Transform): which asserts 2!^n input length (to speed-up).\\
     15͏    [ !^ https://www.ni.com/docs/en-US/bundle/diadem/page/genmaths/genmaths/calc_fouriertransform.htm#d20e91 ]\\
     16͏    DCT (Discrete Cosine Transform) is cut-down variant of DFT that uses only cosine transform; thus of reduced complexity but also minimum resolution:\\
     17͏    DFT: 2 * π / N\\
     18͏    DCT: 2 * π / (2 * N)\\
     19͏    ; somewhat comparable to "full_chroma_int": (not complexity-wise)\\
     20͏    https://github.com/MasterInQuestion/attach/commit/e18e5cd26b9d27bacefa685ba70d0d41bc7d7cbc#diff-75a0130f1814ba26c23e3a6fc6ce5046620163061d392086bc55e2c13c145bf9L264
    1221
    13     FFT (Fast Fourier Transform) is fast implementation of DFT (Discrete Fourier Transform): which asserts 2!^n input length (to speed-up).
    14     [ !^ https://www.ni.com/docs/en-US/bundle/diadem/page/genmaths/genmaths/calc_fouriertransform.htm#d20e91 ]
     22͏    Discrete Cosine Transform, Fourier Transform:\\
     23͏    https://users.dimi.uniud.it/~antonio.dangelo/MMS/2013/lessons/L11lecture.pdf\\
     24͏    https://users.dimi.uniud.it/~antonio.dangelo/MMS/2013/lessons/L03lecture.pdf\\
     25͏    https://www.youtube.com/watch?v=k3byqIaULb8\\
     26͏    https://www.google.com/search?hl=en&gl=ca&num=100&filter=0&q=site:users.dimi.uniud.it/~antonio.dangelo/MMS+inurl:lessons
    1527
    16     DCT (Discrete Cosine Transform) is cut-down variant of DFT that uses only cosine transform; thus of reduced complexity but also minimum resolution:
    17     DFT: 2 * π / N
    18     DCT: 2 * π / (2 * N)
    19     ; somewhat comparable to "full_chroma_int": (not complexity-wise) [https://github.com/MasterInQuestion/attach/commit/e18e5cd26b9d27bacefa685ba70d0d41bc7d7cbc#diff-75a0130f1814ba26c23e3a6fc6ce5046620163061d392086bc55e2c13c145bf9L264]
     28͏    Fourier vs. Wavelet Transform:\\
     29͏    https://builtin.com/data-science/wavelet-transform#main-content\\
     30\\
     31\\
     32͏    This article currently mainly covers video denoise.\\
     33͏    Info on audio denoise shall be added later.\\
     34\\
     35\\
     36͏== Video ==[=#Video]
    2037
    21     Discrete Cosine Transform, Fourier Transform:
    22       - [https://users.dimi.uniud.it/~antonio.dangelo/MMS/2013/lessons/L11lecture.pdf]
    23       - [https://users.dimi.uniud.it/~antonio.dangelo/MMS/2013/lessons/L03lecture.pdf]
    24       - [https://www.youtube.com/watch?v=k3byqIaULb8]
    25       - [https://www.google.com/search?hl=en&gl=ca&num=100&filter=0&q=site:users.dimi.uniud.it/~antonio.dangelo/MMS+inurl:lessons]
     38͏    FFmpeg supports 8 dedicated [1] video denoise filters:\\
     39͏    [https://ffmpeg.org/ffmpeg-filters.html#bm3d ͏"bm3d"] (BM3D: Block-Matching 3D)\\
     40͏    [https://ffmpeg.org/ffmpeg-filters.html#nlmeans ͏"nlmeans"] (NLM: Non-Local Means)\\
     41͏    [https://ffmpeg.org/ffmpeg-filters.html#fftdnoiz ͏"fftdnoiz"] (FFT then frequency clamp)\\
     42͏    [https://ffmpeg.org/ffmpeg-filters.html#dctdnoiz ͏"dctdnoiz"] (DCT)\\
     43͏    [https://ffmpeg.org/ffmpeg-filters.html#hqdn3d ͏"hqdn3d"]\\
     44͏    [https://ffmpeg.org/ffmpeg-filters.html#vaguedenoiser ͏"vaguedenoiser"]\\
     45͏    [https://ffmpeg.org/ffmpeg-filters.html#owdenoise ͏"owdenoise"] (Overcomplete Wavelet)\\
     46͏    [https://ffmpeg.org/ffmpeg-filters.html#atadenoise ͏"atadenoise"] (Adaptive Temporal Averaging)\\
     47[1] Not counting [https://ffmpeg.org/ffmpeg-filters.html#pp ͏"pp"] (much defunct "libpostproc"), general manipulation [https://ffmpeg.org/ffmpeg-filters.html#fftfilt ͏"fftfilt"] alike.\\
     48͏    .\\
     49͏    More info:\\
     50͏    https://stackoverflow.com/questions/74831282#75348463\\
     51͏    https://forum.shotcut.org/t/8631/12\\
     52͏    https://old.reddit.com/r/ffmpeg/comments/a5ejgi?sort=old#ebyev3t
    2653
    27     Fourier vs. Wavelet Transform:
    28       - https://builtin.com/data-science/wavelet-transform#main-content
     54͏    "bm3d" and "nlmeans" would be of most interest: for best quality.\\
     55͏    More details:\\
     56͏    https://stanford.edu/class/ee367/Winter2020/report/zhou_m_report.pdf\\
     57͏    https://webpages.tuni.fi/foi/3D-DFT/BM3DDEN_article.pdf\\
     58͏    https://github.com/gfacciol/bm3d/blob/master/bm3d.cpp\\
     59͏    https://www.ipol.im/pub/art/2011/bcm_nlm/revisions/2011-09-13/bcm_nlm.htm#index1h2\\
     60͏    https://www.dirk-farin.net/projects/nlmeans/\\
     61͏    https://forum.handbrake.fr/viewtopic.php?t=30135#p139077\\
     62͏    https://github.com/HandBrake/HandBrake-docs/blob/master/source/docs/en/latest/technical/filters-summary.markdown#denoise
    2963
    30     This article currently mainly covers video denoise.  Info on audio denoise shall be added later.
     64[ "bm3d"[=#bm3d]\\
     65͏    BM3D pre-filters via Block-Matching for similar blocks:\\
     66͏    Then applies tuned traditional denoise for each.\\
     67͏    .\\
     68͏    Thereby workaround the overdone of applying typical denoise vastly.
    3169
    32 == Video[=#Video]
     70͏    Using denoised input for Block-Matching may somewhat aid similarity branching.\\
     71͏    But using complex algorithm (BM3D, NLM alike) for this phase mostly helps not, besides pointless computation. ]
    3372
    34     FFmpeg supports 8 dedicated [1] video denoise filters:
    35     - [https://ffmpeg.org/ffmpeg-filters.html#bm3d ͏"bm3d"] (BM3D: Block-Matching 3D)
    36     - [https://ffmpeg.org/ffmpeg-filters.html#nlmeans ͏"nlmeans"] (NLM: Non-Local Means)
    37     - [https://ffmpeg.org/ffmpeg-filters.html#fftdnoiz ͏"fftdnoiz"] (FFT then frequency clamp)
    38     - [https://ffmpeg.org/ffmpeg-filters.html#dctdnoiz ͏"dctdnoiz"] (DCT)
    39     - [https://ffmpeg.org/ffmpeg-filters.html#hqdn3d ͏"hqdn3d"]
    40     - [https://ffmpeg.org/ffmpeg-filters.html#vaguedenoiser ͏"vaguedenoiser"]
    41     - [https://ffmpeg.org/ffmpeg-filters.html#owdenoise ͏"owdenoise"] (Overcomplete Wavelet)
    42     - [https://ffmpeg.org/ffmpeg-filters.html#atadenoise ͏"atadenoise"] (Adaptive Temporal Averaging)
     73[ "nlmeans"[=#nlmeans]\\
     74͏    [ Quote <Non-local means denoising for preserving textures>:\\
     75https://scikit-image.org/docs/stable/auto_examples/filters/plot_nonlocal_means.html\\
     76͏    Non-Local Means replaces pixels' value by an average of selected pixels':\\
     77͏    Small patches centered around pixels are compared to the one of interest: with the average performed only when the patches share enough similarity.\\
     78͏    As result, the algorithm may restore well textures that would be otherwise blurred. ]
    4379
    44 [1] Not counting [https://ffmpeg.org/ffmpeg-filters.html#pp ͏"pp"] (much defunct "libpostproc"), general manipulation [https://ffmpeg.org/ffmpeg-filters.html#fftfilt ͏"fftfilt"] alike.
    45     .
    46     More info:
    47       - [https://stackoverflow.com/questions/74831282#75348463]
    48       - [https://forum.shotcut.org/t/8631/12]
    49       - [https://old.reddit.com/r/ffmpeg/comments/a5ejgi?sort=old#ebyev3t]
     80͏    [ Quote <Comparison of CNN based and self-similarity based denoising methods> # Introduction:\\
     81https://stanford.edu/class/ee367/Winter2020/report/zhou_m_report.pdf\\
     82͏    NLM averages pixels with a similar neighborhood.\\
     83͏    Like other traditional methods, NLM can produce a high denoising quality.\\
     84͏    But has the drawback of low computational efficiency; and requires manual choice of some parameters:\\
     85͏    Such as the patch size, search window size, and cut-off distance. ]\\
     86<!^>͏    This is primarily for major NLM implementations seem to be false NLM:\\
     87͏    Not really "Non-Local" thus involve the patch size, search window etc. non-essential.\\
     88͏    [ Quote <Non-Local Means Denoising> # Introduction:\\
     89https://www.ipol.im/pub/art/2011/bcm_nlm/article.pdf\\
     90͏    Since the search for similar pixels will be made in a larger neighborhood, but still locally, the name "non-local" is somewhat misleading. ] ]
    5091
    51     "bm3d" and "nlmeans" would be of most interest: for best quality.
     92[ "favg"[=#favg]\\
     93͏    Hypothetical new filter derived from NLM:\\
     94͏    The NLM related papers didn't well explain why the patch similarity procedure was even necessary.\\
     95͏    Wouldn't just mere averaging all pixels within specific value threshold: repeat for each unprocessed pixel group; adequately work?\\
     96͏    (Full-Image Averaging) [ #11019 ] ]
    5297
    53     More details:
    54     - [https://stanford.edu/class/ee367/Winter2020/report/zhou_m_report.pdf]
    55     - [https://webpages.tuni.fi/foi/3D-DFT/BM3DDEN_article.pdf]
    56     - [https://github.com/gfacciol/bm3d/blob/master/bm3d.cpp]
    57     - [https://www.ipol.im/pub/art/2011/bcm_nlm/revisions/2011-09-13/bcm_nlm.htm#index1h2]
    58     - [https://www.dirk-farin.net/projects/nlmeans/]
    59     - [https://forum.handbrake.fr/viewtopic.php?t=30135#p139077]
    60     - https://github.com/HandBrake/HandBrake-docs/blob/master/source/docs/en/latest/technical/filters-summary.markdown#denoise]
     98͏    Caveat the factual validity of some linked.\\
     99\\
     100\\
     101͏== Audio ==[=#Audio]
    61102
    62 ==== "bm3d"[=#bm3d]
    63     BM3D pre-filters via Block-Matching for similar blocks:
    64     Then applies tuned traditional denoise for each.
    65     .
    66     Thereby workaround the overdone of applying typical denoise vastly.
     103͏    FFmpeg supports ? dedicated audio denoise filters:\\
     104͏    ...\\
     105\\
     106\\
     107͏== See also ==[=#See-also]
    67108
    68     Using denoised input for Block-Matching may somewhat aid similarity branching.
    69     But using complex algorithm (BM3D, NLM alike) for this phase mostly helps not, besides pointless computation. ]
    70 
    71 ==== "nlmeans"[=#nlmeans]
    72     [ Quote <Non-local means denoising for preserving textures>:
    73 https://scikit-image.org/docs/stable/auto_examples/filters/plot_nonlocal_means.html
    74     Non-Local Means replaces pixels' value by an average of selected pixels':
    75     Small patches centered around pixels are compared to the one of interest: with the average performed only when the patches share enough similarity.
    76     As result, the algorithm may restore well textures that would be otherwise blurred. ]
    77 
    78     [ Quote <Comparison of CNN based and self-similarity based denoising methods> # Introduction:
    79 https://stanford.edu/class/ee367/Winter2020/report/zhou_m_report.pdf
    80     NLM averages pixels with a similar neighborhood.
    81     Like other traditional methods, NLM can produce a high denoising quality.
    82     But has the drawback of low computational efficiency; and requires manual choice of some parameters:
    83     Such as the patch size, search window size, and cut-off distance. ]
    84 <!^>    This is primarily for major NLM implementations seem to be false NLM:
    85     Not really "Non-Local" thus involve the patch size, search window etc. non-essential.
    86     [ Quote <Non-Local Means Denoising> # Introduction:
    87 https://www.ipol.im/pub/art/2011/bcm_nlm/article.pdf
    88     Since the search for similar pixels will be made in a larger neighborhood, but still locally, the name "non-local" is somewhat misleading. ] ]
    89 
    90 ==== "favg"[=#favg]
    91     Hypothetical new filter derived from NLM:
    92     The NLM related papers didn't well explain why the patch similarity procedure was even necessary.
    93     Wouldn't just mere averaging all pixels within specific value threshold: repeat for each unprocessed pixel group; adequately work?
    94     (Full-Image Averaging) [ #11019 ] ]
    95 
    96     Caveat the factual validity of some linked.
    97 
    98 
    99 == Audio[=#Audio]
    100 
    101     FFmpeg supports ? dedicated audio denoise filters:
    102     ...
    103 
    104 == See also[=#See-also]
    105 
    106     [https://github.com/FFmpeg/FFmpeg/blob/master/libavfilter/ https://github.com/FFmpeg/FFmpeg/blob/master/libavfilter/][av]f_${Name}.[ch]
    107     E.g. https://github.com/FFmpeg/FFmpeg/blob/master/libavfilter/vf_nlmeans.c
     109͏    [https://github.com/FFmpeg/FFmpeg/blob/master/libavfilter/ https://github.com/FFmpeg/FFmpeg/blob/master/libavfilter/][av]f_${Name}.[ch]\\
     110͏    E.g. https://github.com/FFmpeg/FFmpeg/blob/master/libavfilter/vf_nlmeans.c