Technical Steps to Make Digital Video Seem like Film to Audience - A Short Guide

Ebrahim Saadawi said:
Razz: compression is the devil for Image quality but the angel for practicality. The world wouldn't work without compression and everything would be of much, much lower quality. If one wants to emulate rich negative film raw and uncompressed with high bit depth is a very very essential step in my opinion.
Click to expand...

Certainly true. I don't mind compressed formats for delivery, they are essential especially for the web and the quality can be pretty good for the data rates everything considered. Even a third or fourth gen film release print was significantly degraded from the original negative by the time it had been through conforming, timing, optical FX etc.

I just don't like them for camera original acquisition for the most part, though I do understand the necessity for them sometimes. Computers are getting fast enough and storage cheap enough that compression is not as essential for acquisition as it used to be. Though when I start backing up and transcoding longer form docs like the Max Ginsburg project I've started, my D16 eats 2TB drives for lunch. Elle recently posted some BTS notes from a dramatic feature she participated in as second camera op for Digital Bolex's first grant project for women DP's. Seven day shoot produced 6TB of footage.
 
Last edited:
Well, there is lossy compression and then there is lossless compression.
Still photo cameras have been using lossless compression for years to reduce storage size with no image quality loss, and so do all current Blackmagic cinema cameras. I'd argue that compression actually shines at acquisition, because compression of Bayer data (both lossless and lossy) is much more efficient than compression of debayered images. Decent debayering always includes some sharpening and this hinders any subsequent compression, and this is on top of the 3x increase of data size that debayering does by definition.
 
Is there an easy explanation for how could one discard information, reduce size and still be ''lossless compression"? Seems to be against the laws of the universe.
 
So we can take the 3A2B5C information and recreate the AAABBCCCCC again, with absolutely NO loss? But how does it reduce the size without discarding any information?

Does Lossless compression mean really 100% no loss or just another way of saying ''visually unnoticeable loss"?
 
Ebrahim Saadawi said:
So we can take the 3A2B5C information and recreate the AAABBCCCCC again, with absolutely NO loss? But how does it reduce the size without discarding any information?

Does Lossless compression mean really 100% no loss or just another way of saying ''visually unnoticeable loss"?
Click to expand...

It is mathematically identical. There is no loss.
 
Ebrahim Saadawi said:
So we can take the 3A2B5C information and recreate the AAABBCCCCC again, with absolutely NO loss?
Click to expand...

Yes. Come on dude.


But how does it reduce the size without discarding any information?
Click to expand...

AAABBCCCCC is 10 bytes
3A2B5C is 6 bytes.

Does Lossless compression mean really 100% no loss or just another way of saying ''visually unnoticeable loss"?
Click to expand...

It means absolutely no loss. Not a bit was harmed in the making of the compression.

Open Notepad. Copy this into it. Save it, as something.
Now right-click it. Click Send to, then click Compressed (zipped) folder.
Compare sizes.
Open the zipped version. Compare.
Repeat over and over until you believe.

Still not enough? Google it.
 
cpc said:
Well, there is lossy compression and then there is lossless compression.
Still photo cameras have been using lossless compression for years to reduce storage size with no image quality loss, and so do all current Blackmagic cinema cameras. I'd argue that compression actually shines at acquisition, because compression of Bayer data (both lossless and lossy) is much more efficient than compression of debayered images. Decent debayering always includes some sharpening and this hinders any subsequent compression, and this is on top of the 3x increase of data size that debayering does by definition.
Click to expand...

I have no issue with mathematically lossless compression. Even some types of lossy compression are OK depending on what your final delivery product is. Perfection is relative to purpose. But lossy compression does make choices about what is visually relevant and discards data that does make a visible difference. There is a significant difference when you are grading originals in what can be recovered and manipulated and what can't, especially fine differences in color and texture of deep shadows where data is thin anyway. It can make a huge difference when tweaking temporal noise reduction for underexposed images.

18621854920_b13e5d1186_o.jpg


18155346012_5718e3dfa0_o.jpg
 
Ebrahim Saadawi said:
Ryan: that's a very informative and entertaining post. Would you possible have the time to write something about the history of camera manufacturing (cinema) from film to digital. You seem to know a lot and about Panavision and your writing is very poetic. I'd love to read something long.
Click to expand...

I know a thing or two about Panavision as I very briefly worked there years ago before going into the camera department full time.

Regarding the history of camera manufacturing from film to digital, I know a good deal but unfortunately not enough to do an article justice. It would also be a very big undertaking. All I do know is camera size has nothing to do with the look, nor does it have characteristics of either video or film. For every heavy and big film camera, I can show you a heavy and big video camera. For every small camcorder, I can show you a tiny consumer 8mm film camera.

Regarding look, the weight of the camera can sometimes have an effect. Lightweight cameras have less mass and thus when moving, say handheld, have less momentum. You can sometimes feel this on screen. However, the amount of weight needed for decent mass/momentum isn't something that is only found in video or film. Just a fact though. I can feel the lack of mass in iphone video for example.
 
Last edited:
Disagree on first part, I would love to read an article, and agree on the camera size part.

The thing is, I am trying to analyze the characteristics of films, cinema features, that differ from video shot on smaller ''video'' cameras and TV, and heavy, steady motion is one of the very prominent aspects while shakiness and lightweight phones/camcorders with the handheld aesthetic immediately gives the video look.

Is it tied to celluoid film vs Digital cameras? no, these exist in all sizes, just more like usual cinematic images vs usual video images.

Filmic here in my article isn't as much about actual negative film as it is about usual cinema movies shot on film. Analyzing the technical quality differences between begative film and digital sensors would take a 1000 page book, too many film stocks and too many digital sensors, so I just listed the common ones a begginner can apply, right now, to make his video look filmic.
 
combatentropy said:
Yes. Come on dude

AAABBCCCCC is 10 bytes
3A2B5C is 6 bytes
Click to expand...

Bear with me sorry, it's a common fact physicians aren't good at math.

To take that example, actually I see information lost. The first two As were thrown away and replaced with a number 3 and one B, and 4 Cs are lost. They're gone, so if I want to re-construct the first original form I'd have to guess/interpolate these letters out of nothing based on that number.

And if there's nothing lost, why is it 6bytes then not still 10?

I am sorry I tried googling and reading up but even the simplest explantations aren't simple enough, so a very simple basic explanation of how could one reduce a file size without losing data would be helpful! I never got how companies offer Compressed Raw Data, and other offer Uncompressd' 8 bit 4:2:0 data, what does compression mean here!
 
Ebrahim Saadawi said:
To take that example, actually I see information lost. The first two As were thrown away and replaced with a number 3 and one B, and 4 Cs are lost. They're gone, so if I want to re-construct the first original form I'd have to guess/interpolate these letters out of nothing based on that number.

And if there's nothing lost, why is it 6bytes then not still 10?
Click to expand...

Lossless compression is all about a more effective representation of data. See, there is not one single representation of information in computers. Not at all. If you find some redundancy in the original representation, you can then reduce the size without any loss of information.
How does this work? You get a stream of bytes. Knowing something about the semantics of this stream of bytes, you find a better representation of the information and encode the bytes using this. Then the decoder knowing what the encoder does, decodes the bytes to exactly the original stream. That's all that happens.

Now lets try working with the quoted example. For starters, here is the detailed explanation of the algorithm: https://en.wikipedia.org/wiki/Run-length_encoding

And here is a simplified breakup:
We have the stream of bytes AAABBCCCCC. That's 10 bytes. So how do we reduce the size without losing information?
First, we notice that there are runs of the same consecutive characters. So let's do the following in our encoder: each time we see a run of the same character we will write out the number of times it occurs in this run and what the actual character is.
So our encoder sees three A's and dumps 3A, then sees 2 B's and dumps out 2B, and finally, it sees 5 C's and dumps out 5C.
Or, all the output is 3A2B5C.

So far, so good. We had an input of 10 bytes and we get an output of 6 bytes. Seems like we are compressing data down.
But how does the decoder reconstruct the data? Well, simple enough. The decoder knows exactly how the encoder constructs the compressed stream. So it reads 3A, says to itself "Aha! The encoder saw three consecutive A's here" and outputs AAA, then reads 2B, thinks "ooook, we had two B's in the uncompressed stream" and outputs BB, finally it reads 5C and outputs CCCCC.
Or we get the decoded stream AAABBCCCCC. Incidentally, it is exactly the same as the stream before compression. Hence, lossless compression.

This is a very simple lossless compression scheme, but it is still useful for some tasks and is actually used in real world scenarios.
Also check out this post (below the image) for an extreme compression example using the same algorithm: http://www.dvxuser.com/V6/showthrea...-Files&p=1986544202&viewfull=1#post1986544202

Ebrahim Saadawi said:
I never got how companies offer Compressed Raw Data, and other offer Uncompressd' 8 bit 4:2:0 data, what does compression mean here!
Click to expand...
"Uncompressed 4:2:0" is a bit of an oxymoron, because chroma subsampling itself is already a type of lossy compression. What's meant in this case is that there is no any compression after subsampling.
 
What he said.

Compression is always about finding patterns. If somehow you could make an image without a bit of pattern in it, then it cannot be compressed. This is why video with much pattern in it, like a shot from a tripod of a person in front of a plain wall, will have fewer compression artifacts than a shot with very little pattern in it, like a handheld shot of a cluster of swaying trees, beside a waterfall, behind a rising mist, behind a swarm of mosquitos, behind smoke rising from a campfire.

But lossless compression is possible, because information and data are not the same thing, at least in the strict way that computer scientists use those two words. It's not unlike how one person can convey something in few words, and another talkative fellow blathers on and on, but ends up conveying no further information.
 
So in the AAA example, the first two As are discarded and replaced with one digit, so now we discarded information thus reducing size, but then in the decoding process, the remaining A is cloned/copied two times thus we get the AAA again with the same original size and information?

So this example can only be used if there is a consecutive number of absolutely identical data which you can discard all of them but one then be able to bring back by cloning the remaining one? In that case the data has to be 100% identical otherwise they'd be interpolating and loss happening.

Do I get this right?
 
Yes. It was a simple example to show that lossless compression is possible. Actual lossless compression formulas, like Huffman Coding and Arithmetic Coding, are more complicated. PNG is an example of a lossless image format (mathematically lossless). I've read that it gets an image down to 1/4 to 1/10 its original size. It uses a formula called DEFLATE. You can google each of these formulas, or "algorithms," to try to understand exactly how they work their magic.
 
Ebrahim Saadawi said:
So in the AAA example, the first two As are discarded and replaced with one digit, so now we discarded information thus reducing size, but then in the decoding process, the remaining A is cloned/copied two times thus we get the AAA again with the same original size and information?

So this example can only be used if there is a consecutive number of absolutely identical data which you can discard all of them but one then be able to bring back by cloning the remaining one? In that case the data has to be 100% identical otherwise they'd be interpolating and loss happening.

Do I get this right?
Click to expand...

Yep, lossless compression is all about reducing size by removing data by a process that is 100% reversible. Lossy compression is much more effective in terms of size reduction because it is allowed to calculate what is important to the image and what is not. A good video codec does its best to only discard information that will not be missed when you look at the final image.

If you had a plain black 1920x1080 video you could save the exact RGB value of every single pixel and the file size would be just as big as any other 1920x1080 image of literally anything. An action scene from a movie, a very slow moving scenery shot, a shot of water (which is very complex as everything is changing) would all take up the exact same amount of space.

But if I told the computer that every single pixel of every single frame is just the same 0,0,0 RGB value, I could save tons and tons of space and still end up with the exact same video as when we started. I couldn't use that same method for an action scene, obviously, as it is immensely more complex.

I don't know much about the actual details of lossless and lossy compression algorithms, but they are both phenomenal (lossy probably more so to me) in their ability to save space and still produce incredible images.
 
You must log in or register to reply here.
Back
Top