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

[j1clark@ucsd.edu]

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?

The one idea of 'lossless' compression is to take advantage of the fact that there is redundancy in the image. The example give is that there is a 3 byte pattern, 'AAA', and that the 3 "A"s are redundant, and by encoding them as 3A, you 'squeeze' out that redundancy. It would be more obvious if the pattern were 'AAAAAAAAA', and clearly 9A would compress nine bytes in to 2 bytes.

If the 'expansion' rule is to take the first byte, a number, then replicate the second byte by that number, so to expand a compressed sequence, one would take "9A", and using first byte to replicate the second byte, one has a lossless transform.

The various compression algorithms are more complex, but Run Length Encoding, is one of the most obvious.

In most text files, for a Latin based alphabet say for English, putting aside punctuation, there are only 26 letters, add to that another 26 for 'capital' letters, that is only 52 letters, but the usual text file has a binary representation of 8 bits, and so over 200 values are 'wasted', and can be 'squeezed' out, and still be expanded losslessly.

Again that's sort of a gross simplification, but that's the idea.

I don't recall the exact numbers of these days one can get down to 30-40% lossless compression with 'text' files.

In 'lossy' compression say one were to drop all 'vowels' from the language... most trained 'readers' in a language can adapt to no vowels, and so, one hs lssy cmprsn bt stll 'ndrstndbl... srt 'f... may need some diacritical marks...

 

[Ebrahim Saadawi]

That's very informative. I guess some of our digital video cameras do write imgs tht wy n ordr 2 fit mor in2 za crd

 

[j1clark@ucsd.edu]

That's very informative. I guess some of our digital video cameras do write imgs tht wy n ordr 2 fit mor in2 za crd

Exactly... and of course that means that if a lossy compressed stream of data is hit with a lightening bolt, it may cause significant loss until the next 'full frame'... for encodings that use a 'full frame', 2d compressed image, then for some some number of subsequent frames only compress the differences...

 

[Joshua_G]

Hi Ebrahim,
Thank you for your article.

A question concerning apparent sharpness.
You wrote few times that the video of Canon DSLRs looks very soft, compared to that of some Nikon DSLRs.
Is it due to different sharpening, or due to another technical aspect?

 

[Razz16mm]

Hi Ebrahim,
Thank you for your article.

A question concerning apparent sharpness.
You wrote few times that the video of Canon DSLRs looks very soft, compared to that of some Nikon DSLRs.
Is it due to different sharpening, or due to another technical aspect?

Canon uses an optical low pass filter on many of their DSLR's. Nikon not so much.

 

[Joshua_G]

Canon uses an optical low pass filter on many of their DSLR's. Nikon not so much.

Thanks.
Is the optical low pass filter removable?

 

[Razz16mm]

Thanks.
Is the optical low pass filter removable?

No, it is generally a sealed part of the sensor package. You overcome it by adding sharpening in post. IMO an OLPF does more good than harm, it reduces aliasing artifacts and false detail that that can make an image falsely appear sharper than it really is. The difference is true resolution is very slight.

 

[Greg_E]

No, it is generally a sealed part of the sensor package. You overcome it by adding sharpening in post. IMO an OLPF does more good than harm, it reduces aliasing artifacts and false detail that that can make an image falsely appear sharper than it really is. The difference is true resolution is very slight.

It is also part of the optical design, if you just remove it, the light will focus at a very different point. How different depends on the materials and thickness of the filter pack. Could be up to a few millimeters of difference depending on the specific camera.

 

[Ebrahim Saadawi]

Do you specifically mean Canon vs Nikon DSLRs or in general?


Either way it's a hard question to answer, the elements that make a high resolution camera. It's certainly not one.


First of all the


1- Lens sharpness/resolving power: The first element in creating high resolution images, if this one is not achieved all the other subsequent elements are not beneficial.


2- Anti-Aliasing filter existence: Wether there's a filter that ever-so-slightly blurs the image to avoid aliasing or lack of. This is not as big of an element as the othets, in fact hardly makes a difference in a camera's final resolution power, it's so slight.


3- The sensor: the most important element of all together with the lens. Sensor technology is a very complex subject, but of course a sensor with a higher pixel count produces sharper images than one with a low pixel count.


4- Downsampling method: In cameras that have a higher pixel count than the final output, the camera has to use a way to downscale that high number of pixels into the small delivery bucket, from 24mp to 2mp (1080p) in the case of DSLRs. This step is a very important step in determining the final image resolution of the camera system. Some cameras take the easy way and just skips entire lines of pixels to get to the 1080p, which results in softer images compared to other methods as well as high aliasing and moire (Canon DSLRs below the 7D II use that method of down-sampling). Another way is to group pixels together and perform what's called ''pixel-binning'' which is what the 7D II, 5D III and all the new Nikon DSLRs do, it's a superior way vs line skipping as it reduces aliasing and gives a resolution advantage. The other best option of all, is taking the entire sensor RAW image, at full size, and then downscaling that to the delivery resolution using a high-end downscaling algorithm (ex: Lanczos/Cubic). This gives a very, very high advantage in resolution and also eliminates aliasing issues, but requires a very powerfull processor in order to perform such a heavy task in real-time, the Samsung NX1 uses this method.


5- Processing pipe-line: After that, the imagr needs to get compressed by throwing away data including resolution, and colour. It can be compressed heavily to 8 bit 4:2:0 (DSLRs) or compressed lightly with a 10bit 4:2:2 codec for example, and some camera don't compress at all they give you the full raw uncompressed images with the full resolution and colour (Blackmagic cameras). The liger the compression the higher the resolution and colour fedility you'll get everything else being equal.


6- The Post-processing pipe-line: up to the user, he can make mistakes that result in very low resolution like delivering in interlaces or at SD resolution or he can keep the original full Data.




These are the factors that affect the resolution of the final video image on cameras, if you're asking specifically about which ones Nikon uses vs. Canon to get higher resolution video images, I don't know. But if I had to guess, it's the sensor resolution and the downsampling method.






This was all about resolution, how much detail and information in the image. Sharpness however, is a completely different subject, it's how emphasized the edges of the images are, a technique used to trick the eye into thinking it's higher resolution. What affects that is again the


1 -Processing pipe-line: whether there's digital sharpening added in the pipe-line or not during compression


2 -The in-camera settings: Picture profiles, control, styles, paint menus, and so on, in which you specifiy how much sharpening you want to add. Unfortunately some cameras don't allow for getting absolute zero sharpening, and the sharpening amounts at Zero differ from one company to another, for example an early Canon SLR at 0 sharpness has 100% no digital sharpening while Nikon DSLRs have a slight notch of digital sharpening at zero (and since Sharpness level tricks our eyes into thinking it's higher resolution this is a small element in why Nikon DSLRs are sharper/higher resolution to viewers, but not the main factors (sensor and downsampling methods are). Newer generation Canon DSLRs have the same behavior as in they don't allow achieving zero sharpening, even at zero there's still a slight amount (T6s, Eos M3, 5Ds) and Nikon DSLRs don't allow it too.


3 -The post-processing: The amount of sharpening added in the post-processing/grading step and in the final viewing monitor/TV.




In terms of consumer cameras resolution ranking (only, solely, just, resolution, the amount of information they deliver in video mode, not any other element) are:


1- GH4 & NX1 & BMP4K & XC10 & AX100 at 4K


2- A7s & C100 & 5DIII Raw & BM pocket at 1080p


3-Nikon DSLRs, Sony A6000, Canon's newer gen T6s/m3 .


4-Canon's old gen 5D, 7D, 70D, T5i, etc


5-Fujifilm, olympus and Leica cameras


Again, just on resolution bases nothing else. The No. 1 cameras above reach a very high resolution similar to the highest end cinema cameras available today (Reds, F55, Alexa, Varicam, Cion, FS7, etc) there are slight resolution differences but not enough to up-rank them, but of course only in resolution, these cameras have other enormous advantages in both their image quality and function. The only noticeably higher resolution video cameras than no. 1 cameras are limited to 6K Dragon, 6K Alexa 65, 8K weapon and so on.

 

[Razz16mm]

For a "film like" look resolution is overrated, unless you are trying to emulate 70mm. Best 35mm projected resolution I ever saw in theater off a SMPTE test print was 850 lpph.

 

[Joshua_G]

Hi Ebrahim,
Thank you very much for your very comprehensive and detailed reply.

…
In terms of consumer cameras resolution ranking (only, solely, just, resolution, the amount of information they deliver in video mode, not any other element) are:

1- GH4 & NX1 & BMP4K & XC10 & AX100 at 4K

2- A7s & C100 & 5DIII Raw & BM pocket at 1080p

3-Nikon DSLRs, Sony A6000, Canon's newer gen T6s/m3 .

4-Canon's old gen 5D, 7D, 70D, T5i, etc

5-Fujifilm, olympus and Leica cameras

…

How would you rate the consumer cameras for:

1. 'Filmic' look (including aliasing, rolling shutter, in-camera compression)?
2. Video shooting handling (auto focus and AF tracking, focus peaking, HDMI out)?

(Having very small budget, I'd probably opt for lower than #1 in the list)…

 

[combatentropy]

Best 35mm projected resolution I ever saw in theater off a SMPTE test print was 850 lpph.

Is lpph line-pairs per picture height or just lines per picture height? Because that makes it either 850p (nearly HD) or 1700p (nearly 4K).

 

[Ebrahim Saadawi]

Hi Ebrahim,
Thank you very much for your very comprehensive and detailed reply.




How would you rate the consumer cameras for:

1. 'Filmic' look (including aliasing, rolling shutter, in-camera compression)?
2. Video shooting handling (auto focus and AF tracking, focus peaking, HDMI out)?

(Having very small budget, I'd probably opt for lower than #1 in the list)…

Well isn't that a hard question?

What makes it hard, is that no one camera has all the elements, so you have to give up ones based on what you can give up. Listing each camera's failings and winnings on the market would take a book, not even article. I can help more if you specify a couple of models, a price range, the aimed use, and so on.

 

[Razz16mm]

Is lpph line-pairs per picture height or just lines per picture height? Because that makes it either 850p (nearly HD) or 1700p (nearly 4K).

Lines per picture height. About the same as the best quality well resolved 1080p. The print you see in a theater is generally 4th generation down from the original negative and resolving power is limited by gate weave and jitter in the projector too. It is color and gamma curve as much as anything that gives film its character. Even DR was not nearly what people think for many films. Kodachrome and Ektachrome slide films were both less than 7 stops DR.
Personally I don't think digital cameras look at all like film even at their best. It is a different medium altogether. How one shoots and grades the footage has far more impact on the look than minor differences in cameras.

 

[Joshua_G]

Well isn't that a hard question?

What makes it hard, is that no one camera has all the elements, so you have to give up ones based on what you can give up. Listing each camera's failings and winnings on the market would take a book, not even article. I can help more if you specify a couple of models, a price range, the aimed use, and so on.

Thanks.
Okay, I'll try to simplify, or narrow it down.
How about consumer cameras of up to $2K (body only), between FF, APS-C and micro 4/3, top 3 rows, in the following categories:

1. 'Filmic' look, including colors, or color science.
2. Auto focus and following auto focus performance.
3. DR.
4. Low light performance.

If this is still too hard, the first 2 categories will suffice.

 

[j1clark@ucsd.edu]

Even DR was not nearly what people think for many films. Kodachrome and Ektachrome slide films were both less than 7 stops DR.

Since I only know about still film, B&W and printing papers, 7 stops was the goal, when I was active, which would be the 70s.

Dynamic Range was a phrase never used in my recollection. What was used was 'latitude' and that was used in the following types of questions... 'if i over expose my negative by 2 stops, am I screwed'...

Paper also had grades of ranges, with a #5 being 'harder'... more contrast... less dynamic range, than a #1... #2 was considered 'normal'.

Depending on the material, which I used Tri-X or HP4, nominally rated at ASA (ISO) 400, but I rated 200, one could 'over expose' by 2 stops, cut development time, and still get an 'acceptable' print on a #2 paper. However, over exposure did have some negative attributes, and so, one would take care if one had a controlled environment, like a studio, to keep the range of values to match the capabilities of the negative and ultimate output print.

Even using the Zone System to calibrate the process, one always thought in terms of the 7 stop range of the output print.

Since motion picture processes were far more 'hands off' than still print making, (at the very last I could burn and dodge regions of a print that refused to conform to my desires...), I can only guess that the 7 stop 'projection' print range was the goal for how to expose. (These days 'power windows' would be the burn/dodge equivalent... but mostly unavailable to the motion picture process without significant costs of creating various interneg/postiives and masking.)

In the late 60s and 70s because of the changing asthetics of motion pictures (and stills as well...), one began to see pushed processes, less/available light, yielding rather grainy images, but that, according to the aesthetic, was 'more real' than the smooth, minimal grain, studio films.

 

[Ebrahim Saadawi]

Thanks.
Okay, I'll try to simplify, or narrow it down.
How about consumer cameras of up to $2K (body only), between FF, APS-C and micro 4/3, top 3 rows, in the following categories:

1. 'Filmic' look, including colors, or color science.
2. Auto focus and following auto focus performance.
3. DR.
4. Low light performance.

If this is still too hard, the first 2 categories will suffice.

So want to know which camera is best in each category, regardless of anything else about it? That would be easy for example:

Autofocus: Only usable option is the Canon cameras with Dual Pixel AF (70D, 7D MKII)

DR: Sony A7s in S-Log2

Lowlight performance: Sony A7s

Filmic quality & colour science under 2000$: Nothing makes more filmic images than Blackmagic cameras in 12 bit RAW at under 2000$ (and even higher). The pocket and 2.5K.

So Unfortunately, no one camera has all of them so you have to choose the most important ones for ''you''. Example, A7s has Lowlight and DR yet not colour science and pleasing rendition of colours, GH4 has resolution yet not the DR or lowlight performnce, BM cameras have that distinct filmic look and best colours, 12 bit raw colour, yet don't have the lowlight performance and are a bit buggy in usage and battery life. 7D MKII has the video AF yet doesn't have the resolution and overall image quality of the other modern 4K options. C100 mark one at 2900$ has most, DR, lowlight, AF, colour science, resolution, lack of artefacts, yet lacks the small size of the others and doesn't have a slowmotion option.

It's all a compromise unless you buy an Alexa,

well, actually, even with that you compromise 4K resolution and size and lowlight performance of an A7s!

This is a very general response if you want help choosing a camera system for your specific needs my inbox is open.

 

[Joshua_G]

Hi Ebrahim,
Thank you.

…
It's all a compromise unless you buy an Alexa,
well, actually, even with that you compromise 4K resolution and size and lowlight performance of an A7s!
…

:smile:

…
… A7s has Lowlight and DR yet not colour science and pleasing rendition of colours, …

Without good colors rendition, a camera is useless to me.

… GH4 has resolution yet not the DR or lowlight performnce, …

How bad is its' DR and low light performance compared to the second-best ones?

… BM cameras have that distinct filmic look and best colours, 12 bit raw colour, yet don't have the lowlight performance and are a bit buggy in usage and battery life. …

For documentary, which is what I'll shoot most, BM cameras seem to be the least suitable.

… 7D MKII has the video AF yet doesn't have the resolution and overall image quality of the other modern 4K options. …

Without a good-enough overall image quality, a camera is useless to me.

… C100 mark one at 2900$ has most, DR, lowlight, AF, colour science, resolution, lack of artefacts, yet lacks the small size of the others and doesn't have a slowmotion option. …

I can live with a bigger size and no slow motion, or have slow motion, when needed, with a B camera.
Yet, at this point in time, I cannot afford $2,900. Perhaps best is that I should save towards it.

This is a very general response if you want help choosing a camera system for your specific needs my inbox is open.

Thank you very much again.

 

[j1clark@ucsd.edu]

Filmic quality & colour science under 2000$: Nothing makes more filmic images than Blackmagic cameras in 12 bit RAW at under 2000$ (and even higher). The pocket and 2.5K.

I've had no complaints about my BM Pocket. Sure it may not shoot at ISO 3200, but I'm ok with the 1600 it has, but typically shoot at 800. I've been shooting RAW of late, and the biggest deal there is the price for the high performance SD cards. But I have found for my shooting, one of the Best Buy house named Sandisk cards under the Pixtor label work fine. They may not work for longer than a few minute takes, (like hours...), or under all temperature conditions... but they seem to work for my short takes (under 4-5 minutes). Since I buy the 64 GB versions I only have about 15 minutes of capture time anyway...

Using the free version of Davinci Resolve has been satisfactory as well. I find the color manipulation tools to be better and easier to use than what has been available with Premiere and After Effects. I would still use Premiere and AE for my editing tools, but the color portion of post I'd use Resolve.

Premiere has been recently 'upgraded' to better use the Cinema DNG format of the Blackmagic RAW media. But I still prefer Resolve.

 

[theprestige]

So it sounds like the most cinematic camera you can get is the Sony A7S. Which is my next camera. A A7S with Nikon AIS or Samyang primes.