Log Output

[Ashim]

I'd like to know what does it mean to say that the Viper Filmsteream camera delivers an RGB 4:4:4

10 bit LOG output uncompromised by electronic camera processing.

The query is directed toward the Log output, how does it affect the quality of the images compared to

Linear mode?

[Stephen Williams]

I'd like to know what does it mean to say that the Viper Filmsteream camera delivers an RGB 4:4:4

10 bit LOG output uncompromised by electronic camera processing.

The query is directed toward the Log output, how does it affect the quality of the images compared to

Linear mode?

 

Hi,

 

The pictures are green and unprocessed.

 

Take a look at www.digitalpraxis.com Steve Shaw has written about this.

 

Stephen

[David Mullen ASC]

The terms are a bit odd, "LOG" and "RAW" output for the Viper because I assume the signal off of the sensors has to be processed to become LOG, being naturally more linear-ish.

 

At least, that's what a Sony engineer told me when I asked him why the F950 does not output LOG, which was "what's the difference between outputting LOG versus processing LIN to resemble LOG, since a video camera is naturally LIN anyway?"

[Phil Rhodes]

Hi,

 

A video camera - by virtue of the response curves of the CCDs - is nothing very much like either linear or log by default, although it is nearer linear than anything else.

 

The reason Viper outputs log is to make film people feel warm and fuzzy about it; the reason an F950 outputs linear is because it's easier to deal with.

 

I remain to be convinced that it makes any practical difference, and I think it can be quite unhelpful if you are dealing with the extreme ranges of the image.

 

Phil

[Anna Uio]

If you store the logarithm of the signal, it is then a simple matter to undo it with an exponential in post to recover the original signal exactly (up to quantization error due to the 10-bit integer encoding). Logarithms are THE natural way to capture a big range of scale (from tiny to huge) in a fixed amount of space with equal precision given to each "stop" or doubling of size. Floating point (decimal) numbers in computer are all stored this way.

 

Traditional video cameras take the square root of the levels (more or less) to achieve a similar precision across a wide scale, but this does not scale up to higher and higher ranges. Log scales perfectly, because that's exactly what its for! :)

 

Cheers,

Andrew

 

P.S. I'm not associated with the company that makes the Viper. I just code video processing filters.

Edited December 7, 2007 by Andy Uio

[Phil Rhodes]

> Logarithms are THE natural way to capture a big range of scale

 

Why?

 

People keep telling me this, and I'm perfectly willing to be convinced, but just repeating that it's THE way to do it isn't actually very informative.

 

Phil

[adam berk]

To the best of my understanding, 10bit log is basically a way to store more information in a smaller amount of space...a form of compression if you will. I've been told that to store the amount of exposure info contained in 10log, in a linear image, one needs at least 16bits (versus 10). Also, 10log maintains compatibility with current video infrastructure technologies such as hd-sdi.

 

I'm just a n00b with this stuff as well...so someone please correct me if I'm wrong.

 

thanks,

adam

[Anna Uio]

> Logarithms are THE natural way to capture a big range of scale

 

Why?

 

People keep telling me this, and I'm perfectly willing to be convinced, but just repeating that it's THE way to do it isn't actually very informative.

 

Phil

 

Mathematicians & engineers have been using logarithms for hundreds of years. Slide rules used logarithms for calculations before computers. Every high school student learns about them, although it is hard to understand what they are useful for at that age :) There are probably lots of web pages on line that explain why logarithms are useful in diverse fields whenever large ranges of scale are to be considered. A start for anyone who wants to learn about them would be to google "logarithm". Also the link http://en.wikipedia.org/wiki/Logarithm, but that may be a little too mathematical for some.

 

As Video Editing software is moving to 32-bit floating point representation of numbers, instead of 8, 10 ot 16 bit integer, the need for logarithms or square roots is beginning to fade away, since floating point numbers have the "log" thing built into them in a way that you don't see it and just forget its there.

 

Oh, and people keep referring to regular video as linear, but its not. A power function, x^0.45 (which is about the same as square root) is used to compress the range of values. This is pretty wonky and won't scale up as devices become more precise and have larger dynamic range/latitude.

 

Finally, 10bits can be used to store 2^10 distinct values, regardless of where they are distributed logarithmically or using square root. Logarithms naturally allocate the same level of number of value across each "stop" or doubling of intensity. Power functions don't do that, so some stops get allocated more precision than others. Also, it is not really true that logs allow you to store a 16 bit number in 10 bits. Rather, logs allow you to store an any-bit bit number (8-bit, 16-bit, 32-bit, 128-bit) in a smaller number of bits by discarding precision evenly across stops, as opposed to keeping precision in the larger numbers and discarding it in small numbers as simple truncation would do.

 

Cheers,

A

[Anna Uio]

This link might be useful too:

 

http://en.wikipedia.org/wiki/Logarithmic_scale

 

Cheers,

A

[John Sprung]

The query is directed toward the Log output, how does it affect the quality of the images compared to Linear mode?

Log vs linear has to do with how we transport and store image data, not with how the camera captures it.

 

Suppose we have 10 bits available per sample. That gives us 1024 possible values in theory (in practice, there are a few reserved for other purposes). Now imagine that we have a neutral gray card, and we light it so as to have one footcandle at one end of the card, ramping up smoothly to 1000 footcandles at the other end. If we were to assign the numbers that come out of the camera so as to be equal to the footcandles, that would be linear coding. It's simple, and it works.

 

But given that we want to use our combinations of bits as efficiently as possible, is it a good thing? Not really, because of how the human visual system works. Using split screen brightness comparison tests, it has been shown that the smallest difference that the average person can detect is about 1%. (That's not absolutely perfect over the whole range, but it's close.) So, the difference between 100 footcandles and 101 footcandles looks about as big as the difference between 1000 and 1010 footcandles. It's called a JND for Just Noticeable Difference.

 

Thinking back to our imaginary gray card, we're wasting code values by keeping track of the difference between 998 footcandles and 997 footcandles. Worse, we're lumping together everything between 6.5 and 7.5 footcandles, and putting an artificial sharp line between that and the stuff between 7.5 and 8.5. The difference is well over ten JND's, and we see severe banding or posterization.

 

Logarithmic coding spreads the bits around closer to the way the eye works -- the wasted bits from the high end are freed up to give us more subtlety in the shadows.

 

 

 

 

 

-- J.S.

[David Mullen ASC]

Well, someday I'm sure converting digital camera signals to 10-bit LOG for recording and post will become less and less common as recording higher and higher bit rates becomes easier. For example the Dalsa sends out an uncompressed 16-bit 4K RAW Bayer files -- the idea is a good one, but the practicality of uncompressed 16-bit at that resolution and frame rate, etc. is only slowly becoming a reality.

[Phil Rhodes]

In response to earlier answers - yes, I understand what a logarithmic representation is and why it's useful for representing wide ranges. But the image off a video camera isn't a wide range; it's also not linear to begin with, even before any DSP munging that goes on: the chips aren't anything like linear.

 

So what you end up with is a lograithmic representation of... something ill-defined. And while this is true, as it will be for the forseeable future, you're not working in log exposure units or anything of that sort. You're working in "what looks nice". And defining what a logarithm is does not explain why it helps things look nicer.

 

Phil

[Mitch Gross]

In essence, Phil is right. There's no real difference; it's just about the efficiency of storing information.

 

The Phantom HD captures in 14-bit linear RAW. This is how the information is stored. But with the new CineMag flash memory packs, it would be ungainly at best to stack linear files of material shot at different frame rates, resolutions, bit-depth (it is selectable) and a host of other parameters. There would be a lot of re-interpreting of the numerical structure just for the computer to keep up and it would bog down the system. So the 14-bit (or less) linear files are all translated into Log 10-bit for storage on the CineMag, and then put back into linear space on output. And absolutely nothing is lost along the way -- it is mathematically identical.

[Anna Uio]

> But the image off a video camera isn't a wide range

 

Not true. The range of video cameras keeps getting wider as we move to more precise analog to digital converters. Currently the precision of the A/D converters is exceeding what can be represented in standard storage formats. Hence the move to more bits per pixel and logs.

 

There is a huge reason to capture store a wide range or latitude. You can recover/use all the detail in the shadows and highlights that is lost in normal range film or video. If the range is wide enough, you can just forget about exposure settings during shooting and do it later in post.

 

Photographers talk about high dynamic range all the time. For decades there have been techniques for taking multiple low dynamic range shots of the same scene at different exposures and combining them to get a single high dynamic range image. Now they can do it with just a single shot in todays digital cameras.

 

I can't wait for prosumer video cameras to become like the viper. I've coded lots of effects video filters that work ok with standard range images but will benefit immensely from increased precision and range.

 

Also, people keep saying that the CCDs in a camera aren't linear. But that doesn't matter. The software inside the cameras knows about this non-linearity and straightens it out in the stored image.

 

Cheers,

A

[John Sprung]

Well, someday I'm sure converting digital camera signals to 10-bit LOG for recording and post will become less and less common as recording higher and higher bit rates becomes easier. ....

It'll always be a matter of bandwidth/storage in the middle vs. processing power on both ends. So, I'd expect to see both in use well into the future, depending on what the constraints of each particular application are.

 

 

 

 

-- J.S.