Banding in Red Footage

[Dj Joofa]

Joofa
Forgive my ignorance..

Hi Michael,

You don't have to say that. It is by talking and asking questions we collaboratively learn.

Imagine I have a 10bit files that when displayed on a 10bit monitor does not band. Truncating this file to 8bit causes banding. If i take this file and run it through a dithering plugin like the one offered free from http://krokodove.com/
and dither to 8bits. The edges where I would have seen banding have been spread (in a noise like pattern). Is this not dithering?

I would like to have the correct terminology so any help would be welcome?

The term "dither" has historically been understood as the process done to reduce quantization effects on analog signals for transmission over digital channels. In one of the first publications on this topic it was described as:

"In an attempt to reduce the bandwidth required to transmit PCM (Pulse Code Modulated) television, a simple but effective coding technique has been developed which eliminates the contours and intensity steps characteristic of coarsely quantized pictures."

The original intent of the dithering process was to reduce perception of harmonic distortion resulting from quantization inherent in digitizing data. However, for sometime the term is being used for removing banding-type noise in digital images, which have already been quantized, thanks to software based noise reduction software, who don't seem to realize the origins of the term.

You are safe to use the term "dither" these days as hiding banding in digital data via software as this is how most people understand it now.

Though, it must be understood, since the data data has already been quantized before adding "dithering" noise, therefore, this noise is not going to reduce the harmonic distortion effects and is only going to "hide" it. That is why I mentioned in my earlier post that greater amounts of noise will be needed to hide banding artifacts in digital data vs. the amount dither noise added as analog noise before quantization

 

[Michael Lindsay]

Thanks Joofa

I apologised for my ignorance because my question was genuine (not combative). Also the science behind the images is not my area of expertise.

I sometimes grow tired of both the lack of humility and plain 'know it all stupidity' present on lists like this. Displaying some humility when asking somebody something that falls in their (and not mine) field of expertise is hopefully no bad thing?

kind regards

Michael

 

[Michael Lindsay]

Thanks Joofa

One more question..

If I have banding in a 8bit image and add noise I will eventually cover it. If I however start with a higher bit depth and 'intelligently smear/spread (dither down?) at the edges (that will be created by the move to a lower bit depth) surely this is a less noisey process?

And can dithering only be applied, in the digital domain, (and I realise this is a recent use of the term anyway) to a move from a higher to lover bit depths?

Is the move from higher to lower bit depths not analogous to the AD stage?

kind regards

Michael

 

[Daniel Browning]

Dithering is an area where analog gain helps, and unfortunately, Red does not have analog gain.

If there truly was any posterization in the R3D, analog gain would have helped, yes, but no one has demonstrated that.

I have never seen posterization in raw files from any camera: this is because photon shot noise, read noise, PRNU, etc. always provide enough dithering by themselves to suppress posterization and even any measurable amount of quantization error. If posterization did exist in the R3D and it wasn't due to post-processing, I would sooner suspect an issue with the compression than the sensor/electronics.

 

[Graeme Nattress]

I've never seen banding in the actual R3D data itself. I've seen it when photoshop has been telling fibs about how the image looks, I've seen it on 6bit laptop screens, in web quality movies, in quicktime movies etc., but never in the R3D itself.

Graeme

 

[Dj Joofa]

Is the move from higher to lower bit depths not analogous to the AD stage?

This is an important question. In theory one can recover the probability density function of original image from the probability mass of the digital image using a technique that involves among other things Shannon sampling theorem applied to characteristic function (http://en.wikipedia.org/wiki/Characteristic_function_(probability_theory)), and then add noise and requantize to a lower bit depth than originally intended. Please note you can't recover the actual analog signal values (they have been quantized in the digital image, I am talking about the probability densities here, that may be recovered.)

Now if you operate in the digital domain and add noise then you have to see that the statistics you are getting are the same as the process mentioned above. I did not do a formal comparison, however, on the surface it appears to me that the digital process involves one more convolution term, but as I said, I just glanced over the problem, and I have to verify their equivalence or non-equivalence.

May be we should assign that as a homework problem to Daniel Browning. :w00t:

If there truly was any posterization in the R3D, analog gain would have helped, yes, but no one has demonstrated that.

Hi Daniel, since your betrayal I have decided that I am not going to respond to you for about a week, unless of course, you solve the above problem that we want you to solve.

I've never seen banding in the actual R3D data itself. I've seen it when photoshop has been telling fibs about how the image looks, I've seen it on 6bit laptop screens, in web quality movies, in quicktime movies etc., but never in the R3D itself.

Graeme

Graeme, you are treating your own product with kids glove. Kindly send a Red camera to us and we shall knock the living daylights out of it. :w00t:

 

[Andrew M.]

I've never seen banding in the actual R3D data itself. I've seen it when photoshop has been telling fibs about how the image looks, I've seen it on 6bit laptop screens, in web quality movies, in quicktime movies etc., but never in the R3D itself.

Graeme

I did push this 16 bit tiff file to 100% contrast and what it looks like banding in the right lower corner is actually reflection shadow of the leg of the person walking up the stairs.

Graeme is right, I never have seen banding in the 16 bit tiff file as well.
Taking this tiff and converting it to 8 bit JPG it produces clear banding.
What I do not understand is, why when I switch tiff from 16 bit to 8 bit in Photoshop nothing is changing on the screen. Even saving it in 8 bit and reopening will not change anything.

Is the PS so good in converting 16 bit TIFF to 8 bit TIFF that we can’t see the difference and converting to anything else is such a trouble?

 

[Uli Plank]

PS is dithering if you don't tell it not to do that.

 

[Alex Carr]

Were HMI's used in the shot?

 

[Macros]

Just spoke to the DOP and yes, they did use HMI's why you ask?

PS - Yeah its the shadow but as soon as we downscale to HD it appears like banding. We found a work around and will post it here soon for anyone interested. Just doing some more tests

 

[Dustin Bowlin]

the ongoing banding issue

the ongoing banding issue

Hi all,

This thread kind of died, just wondering if there was any follow up to the HMI question. Do HMI's pose any real issue for creating banding? My company shoots a lot of material on a white background with HMIs that fall off with gradients, and we have noticed a lot of banding as well.

We use Cinema Displays as well, and have not encountered any major issues EXCEPT when grayscale type gradients are created. Even in Redcine, there is the appearance of noise and antialiasing, which is only made worse by altering the curves etc.

when output from redcine as 16 bit tiffs it is there etc...

 

[Andrew M.]

Yes, I was wondering if anybody has more experience with banding.
Pain in the neck when converting 4K to HD BR
Is dithering the only way to deal with banding?

 

[Miguel "Macgregor" De Olaso]

I've never seen banding in the actual R3D data itself. I've seen it when photoshop has been telling fibs about how the image looks, I've seen it on 6bit laptop screens, in web quality movies, in quicktime movies etc., but never in the R3D itself.

Graeme

I have seen posterization in the darkest shadows inside Redcine and Scratch, even before exporting to DPX. It usually comes when you have noise and you are trying to use as much DR as possible. I can upload you some footage that shows this problem. At first i thought it was a Scratch problem, but after seeing the issue with different footage from RED ONEs, i concluded it was on the R3D itself. It´s not a HUGE issue, but it´s problematic sometimes.

 

[Graeme Nattress]

Macgregor, you'll have to send me an R3D to look at.

Graeme

 

[Miguel "Macgregor" De Olaso]

PM sent.

 

[Uli Plank]

I've seen most banding after encoding to MPEG-2 or H.264 coming from the fact that RED footage is too clean! Add a tad of grain (well, convincing noise).

 

[Graeme Nattress]

Yes Uli, that's where I've seen banding. That's because such compression schemes can't encode noise unless the bit-rate is stupidly high, so if you have a low noise source image, that noise is ignored, leaving what are essentially flat areas or gradients, and they do, indeed band. That's pretty much the nature of the beast with those codecs and how they work. Your solution is a good one - add just enough noise to force the compression to encode it, and up the bit-rate to compensate.

Graeme

 

[Mark Pugh]

Your solution is a good one - add just enough noise to force the compression to encode it, and up the bit-rate to compensate.

Graeme

Add noise using what, and at what stage?
If the banding is occuring when transcoding from R3D, how can noise be added using Red Alert/Rushes?

 

[Graeme Nattress]

If you transcode to an uncompressed format, then I don't see any banding. The real place you're going to see any is on making a delivery version of the final product, ie h264 for the web, or mpeg2 for DVD or whatever.

Graeme

 

[Simon Blackledge]

I've seen banding going to uncompressed from a r3d in fcp.

transcoded to same (uncompressed 10bit) via AE and the banding was alot less.

Could this be a fcp rgb>yuv issue?


s