New Filters: Conditional Temporal Median Filter and DeSpot filter

[morsa]

I know Ia sk for this things all the time, please forgive.It´s just that I find all these amazing filters and I hope we could add them to Avisynth collection.

It seems that this is what we´ve been looking for since a long time.
For your (our) pleasure.

There are two filters for The Gimp and now for the Cinepaint project that are really interesting.They just need an experienced coder to port them to Vdub or Avisynth.
One is a deconvolution one and the other...Wow is an scratch removal filter (at least it says that).

http://sourceforge.net/project/showf...group_id=75029

There is a port of a Deconv filter from Gimp to Vdub here

http://fcchandler.home.comcast.net/Refocus_YUV_Test.zip

Hoping to find someone kind enough to port them,
Morsa.

[Fizick]

to Morsa:
Thanks, it is interesting.
Cinepaint is free similar Ulead Videopaint?
However "sratch removal" is not filter, but mostly various information. The only source code included is for spatial method, not temporal.

[morsa]

Well the idea now would be to have a detector for the scratchs.Isn't it?.I believe also that this Spatial interpolation would be very good for Logo removal and Subtitles removal.

[Fizick]

I release a new version 0.934 of the filter -
with luma correction and local blur of deleted spots borders.
The filter may be found at:
http://bag.hotmail.ru

COMBINED USAGE

Very good results are may be obtained with combined this filter
with motion estimation and compensation, even if only general motion,
produced my (Fizick) GENMOTION plugin.

[Fizick]

I released a new version of the filter.
Changes 1.0 from 0.934 (Dec 30, 2003)
Added "tsmooth" parameter for temporal smoothing in static areas.
Changed filter name to DeSpot as more appropriate, filter file name to despot.dll,
and function names:
ConditionalDenoise to DeSpot,
ConditionalDenoiseMark to DeSpotMark,
ConditionalDenoiseMap to DeSpotMap,
ConditionalMedian to DeSpotMedian,
ConditionalMedianMap to DeSpotMedianMap.

Corrected info.

The filter may be found at:
http://bag.hotmail.ru

[brett]

So, is there going to be any more work done on this filter? It seems very nice for VHS captures, but it's pretty unusable in its current state.

For example, my heaviest filtering script is:

Convolution3d(c,0,32,128,16,64,10,0)
ConvertToYV12()
Cnr2("xxx",4,5,255)
UnDot()
STMedianFilter(8,32,0,0)
QMF()
BilinearResize(352,208)
UnDot()
Temporalsoften(2,3,3,mode=2,scenechange=6)
Limiter(min_luma=16)

I use it for noisy VHS and cable TV captures. This encodes at 7.33 FPS on an Athlon XP 1600, so that's 3 hours to encode your agerage 1-hour show with commercials taken out.

If I simply add DeSpot to the above script, it's 13 times slower. It goes 0.57 FPS, and takes 37 hours to encode a 45-min show. It would take 3 days to encode your average VHS movie, so it's really not practical.

Maybe there is a similar filter that does well removing the white horizontal lines of noise that show up in old VHS tapes?

[Fizick]

What is your script for DeSpot?
Its speed is depended from parameters setting.

[morsa]

@Fizick

In case this is useful for anything,

http://www.mpi-sb.mpg.de/~hitoshi/re...on/index.shtml

If someone is interested and if this could be usefull here is another link with some examples and a description of the algorithm in pseudo code.


http://www.cs.berkeley.edu/~efros/re...frosLeung.html

[krieger2005]

@Fizick

Thanks for your Filter. It do a good job when you tune them enogh.... But this is the problem: Tuning this filter for a movie is a horrible thing. Not at least because some desciptions of your filter-usage seems to be false

1. you wrote "p1 > mthres > p2". But when im using "p1 > mthres < p2 AND p1>p2" then DeSpotMark show me that some areas were spots.

2.
"sign (default 0)
Set mode for removing of only black or white spots or both:
sign=0 - any spots and outliers (default)
sign=1 - only black (dark) spots and outliers
sign=2 - only white (light) spots and outliers
sign=3 - only black (dark) spots, any outliers
sign=4 - only white (light) spots, any outliers "
But you can only select -2, -1, 0, 1, 2.
When you select 1 or 2 for searching for white/black spots then the filter find dark/light spots too. Why?

3. The Variables
mwidth (default 7)
mheight (default 5)
mp (default mwidth*mheight/3)
seems for some peaple unclear (for me too). Can You explain this a little bit more?

and then technical detail: Avisynth 2.54 does not like your Filter (DeScratch also). When you set it for loading (LoadPlugin) it sait: Could not load ...

I try always to tune the filter for one pic (first). For that i set p2=0 and p1="something" (mthres=p1-1). When the spot where found i tune p2 (+1) or mthres (-1), since the spot were still found by the filter. This method is a "need many time" method and does not give me the results which i estimate -> Artifacts.
For now i do not understand clearly what the filter do so it is difficult to tune the vaiables.
As long, as one can not see which areas were removed by which step of your algorithm it is difficult to tune the values. Something that you could do is to show different steps in different colours: mark_v_forP1, mark_v_forP2...
Then: Can you do that the Mark is transparent (highlighted in the selected colour) so one does not need StackVertical/Horizontal, which does not help, when you have a picturesize of 720/576 (because 720*2=1440, 576*2=1052 an i have the maximum resolution of 1024/768).

But at all the filter does helped me to restorate some movie which were very old and for that

[Fizick]

1. no comment yet.
2. Yes, its my mistake in info.
Correct:
// sign=1 - only black (dark) spots and outliers
// sign=-1 - only white (light) spots and outliers
// sign=2 - only black (dark) spots, any outliers
// sign=-2 - only white (light) spots, any outliers
// sign=0 - any spots and outliers (default)

3. It is mostly Kevin code.
MP is not very clear for me too. It is some limit number of moving pixels.

4. I use v.2.53. Problem was discussed here. Wait.

5. Try get pixel values (and differences) with some Photo-editor.
It is not very robust filter, yes.

6. I have not time now. Plese, wait.

[Fizick]

O.K. I return to it.
New experimental version 1.2 of DeSpot plugin (June 01, 2004):
Added parameters Ranked, weak, maxpts, p1percent,
change mark mode to color spot with weak motion map.

If "ranked" parameter is true, the 3 neighbors in previous frame and 3 neighbors in next frame are ranked (ordered by value), and those min and max are used for current difference calculation

If "weak" parameter is used, the some small weak spots are rejeñted also.
If "numpts" parameter is set, the big spots are rejected also.
If "p1percent" parameter is set, then not strong (by criterion P1) spots(which have many outliers) are rejected also.

The color (pink, green or grey) of noise spot now is depend from parity of mark_v.
Weak motion map is also shown now.

Source now is compatible with MS VC6, VC7 (free tools). But I can not port it from avithynth_c to pure avisynth plugin (needed in some help)

To moderator:
May you add to this thread subject some words:
... and DeSpot filter.

[Fizick]

Now I have a more new version.
Changes 1.3 (June 7, 2004) from 1.2:
Remove parameter "weak" ( "p1percent" is seems more useful).
Add parameter "Dilate" to enlarge spots sizes.
Change parameters order to more functional.
Change default values of some parameters to more optimal.
Update doc.

If "Dilate" mode, spots are enlarged (with range value as the parameter is set) to cover its non-contrast edges and close small gaps between its, by applying a morphological dilate operation to noise (spots) map.

[krieger2005]

Hi,

thanks for development of this filter...
1. Can you put the Link to Despot in the first Post? I must search it always.
- EDIT - I Note, that you does not open the thread... forget this point...
2. I tried DeSpot 1.2 with Avisynth 2.55 alpha using "avisynth_c.dll". It was all okey...
3. Your Site is not updated?

thx
greets krieger2005

[Leak]

Originally posted by krieger2005
1. Can you put the Link to Despot in the first Post? I must search it always.

A small hint: click on the "www" button below Fizick's posts to go straight to his homepage...

np: Gescom - Keynell (Mix 1) (Autechre - Keynell Mixes)

[Fizick]

Originally posted by krieger2005
3. Your Site is not updated?

I have v1.3, but I did not release it
I found a small bug.
And I want to do once more update.
Please wait.

I also wait some responce from users about possible bugs and suggestions.

If nobody responded, why I will release new version?

[Backwoods]

Originally posted by Fizick
If nobody responded, why I will release new version?

Doesn't mean there aren't people using it

[Fizick]

O.K. my site is updated with 1.3.

I wait some respond how useful are new options for various sources.

Now I also think about porting DeSpot to native Avisynth plugin (not C).
Anybody will be interested?

[Fizick]

New DeSpot version 2.0.
It is major release update (probably alpha with bugs):

Main interface code is rewrited, and now filter is native Avisynth plugin (not C-plugin).

Add parameter "median" instead of DeSpotMedian function,
Add parameter "show" instead of DeSpotMark, DeSpotMap, DeSpotMedianMap functions.
Replace parameter "mp" to parameter "merode" (relative)
Change "p1percent" to integer.
Update doc.

[iradic]

hi...

www button is ok... but you could add url in your signature (direct link )

bye

[Fizick]

You are seems too lazy...

http://bag.hotmail.ru