Curiosities

[State of Mind]

What is noise? Can it be completely removed from a DV file? Is it possible to de-interlace without any artifacts left behind? Is it better for the qaulity of my DV files if I use AVISynth and the MS Codec or my Panasonic DV Codec and VirtualDub? What are RGB, YUV2, etc?

[nicco]

Take a look at the sticky threads;)

[State of Mind]

I looked at all the guides and the FAQ done by Sticky and didn't find anything relating directly to noise levels in the DV file. Can it be completely removed altogether using filters in VirtualDub or AVISynth? Like I said, I am also curious of whether it benefits the quality of the DV file to use my Panasonic DV Codec & VirtualDub or the MS Codec & AVISynth? Also, about de-interlacing without artifacts left behind? Are there less artifacts left behind with a higher bitrate? Also, what is RGB, YUV1, 2, etc?
Knowledgeable answers would be much appreciated.

Jeremy DVC

[bb]

Noise is the unwanted alteration of pixel colour values due to tolerances in measuring the intensity of the light the CCD(s) in your camcorder receive(s). So you may call noise an error of measurement. It is mostly apparent when the signal gets low, i.e. in low light conditions.

To get rid of the noise you may use various filters such as Convolution3D, Undot, Deen, and many more. Each of these filters has its strengths, and you'll find much discussion about this topic here. Filtering may reduce noise, but never get completely rid of it, and the downside is, that you may lose sharpness, i.e. the picture may get blurrier.

Noise - although unwanted - eats bitrate, because the encoder will likely see the "dancing spots" as false movement, which is hard to encode because of its irregularity. Thus noise reduction is recommended to get better video at the same bitrate, or smaller filesize for the same quality.

Encoding with higher bitrates reproduces the signal better than lower bitrates, provided that we talk about the same encoder and codec. Noise and interlacing combs are reproduced better, too. That means with higher bitrate there are less artefacts.

Panasonic DV codec vs. MS DV DirectShow filter should produce exactly the same picture, because they are in fact the same code, except for the VfW vs. DirectShow additions (as far as I know).

RGB, YUV, YUY2, YV12, etc. are so-called colourspaces, i.e. the abbreviations describe the way colour information is stored in your video. In brief, the RGB variants store a red, green, and blue value per pixel, whereas the YUV variants store a luminance (= grey brightness) and two colour difference values for one or more pixels.

There's a lot to say about interlacing/deinterlacing - I hope you don't mind searching :) Usually deinterlacing can't be done without artefacts. In spite of that it is still recommended in certain scenarios, e.g. for low bitrate encodings, for progressive display playback, or for encoding using codecs without interlacing support.

bb

[State of Mind]

Thank you, BB. Now THAT was a reply. You literally brought a smile to my face. :D
I am now :confused:wondering:confused: mainly about what I read regarding putting a DV file through so many passes (encodings) that you lose a little bit of information/color/detail everytime? I was wondering what sort of procedure to follow that requires the minimal amount of passes, reducing the amount of inf/col/det lost in a DV file?

PS: I discovered only moments ago that my DVC supports Progressive Recording. I have always done things beginning with an Interlaced source anyway, so I don't think I will change that setting.

Jeremy DVC

[Guest]

Of course it is possible to completely remove all noise. This script will do it:

Tweak(cont=0)

It's slow, this one is much faster:

BlankClip(clip)

The downside, of course, is that they completely remove the signal too. Perhaps that is acceptable for your application.

Joking aside, no filter can perfectly distinguish signal from noise. And you can't even define the difference either!

[bb]

:D

I must have overlooked these outstanding filtering method you described above :)

But you're wrong saying that it's not possible to distinguish signal from noise. Here's a small pseudo-code snippet, which you may want to try to implement:
If <pixel is equal to real world colour> Then
bIsNoise = False
Else
bIsNoise = True
End If

bb

[Guest]

@bb

There seems to be an undefined variable in your function: real world color. Could you provide the header file for that? Thanks.

[State of Mind]

So, removing noise decreases less filesize but keeps the same quality and causes a bit of blurryness as well?
Also, is there a visually-noticable difference in the quality between two videos, one having gone through more passes then the other? Because I read that each pass causes loss of info/qual.

Jeremy DVC

[bb]

Originally posted by State of Mind
So, removing noise decreases less filesize
:confused: If it "decreases less filesize", does that mean it increases filesize?
Originally posted by State of Mind
but keeps the same quality and causes a bit of blurryness as well?
Also, is there a visually-noticable difference in the quality between two videos, one having gone through more passes then the other? Because I read that each pass causes loss of info/qual.

Jeremy DVC
Quality may be reduced by filtering noise, because detail may be lost. But the overall quality can be increased, if you need to encode the noisy source to a given filesize, because denoised video compresses better - thus you get better quality for the bitrate. What you lose by filtering is typically much less than what you'd lose by compressing the noisier unfiltered source.

bb