Premiere Pro Editing – Scaling, Resizing, Zooming and Resolution Loss of Footage Clips

[Tero Ahlfors]

All of that depends on the footage you are scaling. If it’s nice, clean uncompressed footage it will probably scale better than compressed, blocky webvideo.

[Devinda Fernando]

I’m shooting using a Canon 5D MK III on the 1920×1080 ALL-I setting. The codec is H.264. in the .mov or mp4 container.

Does anyone know how the scaling/zooming is handled by the software? i.e – if for example the zoom is 100% then am I to assume that the software doubles the number of pixels of the frame: 1080 x 1920 = 2 Million Pixels (approx) and thus a 100% scale would double the pixels to 4 Million Pixels and therefore resolution would be reduced to 50% of the original?

Am I making any sense on this?

[Tero Ahlfors]

If you scale full HD up by 100 percent it actually quadruples the pixel count to 3840×2160.

[Devinda Fernando]

*** If you scale full HD up by 100 percent it actually quadruples the pixel count to 3840×2160. ***

No it does not… you are confusing 4K footage (which is 3840×2160 pixels) with 2K being scaled by 100%…

[Tero Ahlfors]

If you make a UHD timeline and insert HD footage the footage will fit to the comp when it’s scaled up 100%.

1920*1080=2073600
3840*2160=8294400

8294400/2073600=4

[Blaise Douros]

It all depends on how you’re mastering. Are you going out to 1080P? Then be really careful–that 10-15% number is a bit high, in my completely unscientific and unsupported personal opinion, I wouldn’t go any higher than 10%, otherwise it gets pretty noticeable.

Are you mastering to 720P? Then you’ve got a little room to play around. If mastering to 720P, theoretically you can zoom in to 150% and not lose resolution. The key to this is to bring your 1080P footage into a 720P timeline–you have to reduce it in size by 66% to fill the screen, but if you don’t, you’ll already be zoomed in at the max resolution, since the edges of the footage will be outside of your canvas. So: if you’re scaling up in a 1080P timeline, and mastering to 720P, scale the footage up to 150%. If you’re putting 1080P in a 720P timeline, it should come in already zoomed in, or you can reduce it down to a maximum of 66%.

Mastering to 480P SD? You can zoom in by like, 200, 250% easily.

Mastering to a 360×270 animated GIF? It’s your birthday, and you should thank Eddie the Editor, God of Post Production, for his many blessings. Zoom in as far as you damn well want.

Now, if the question you’re asking is “can I uprez 1080P footage to 4K without losing detail,” the answer is, objectively, no. However, you have to sit reeeeeeally close to a big monitor to see the difference between 4K and HD–I believe Fast and Furious 6, or whatever the latest installment was, used DSLRs as crashcams and mixed quick shots into the regular 4K stuff. As long as you don’t linger on the uprezzed stuff, it should be OK.

[Devinda Fernando]

Ah Ok… I see what you are saying… the Adobe Premiere’s Software calibration of “100%” means that it actually quadruples the size of the image… (like with 720p which is approx 900K pixels when scaled up to 1080 is said to be a 50% increase).

I was referring to the actual standard definition of up-scaling to 100% (as in, doubling the size).

[Devinda Fernando]

Hi Blaise,

I am outputting to the same size (1080) that it was taken on the cam, so just wanted to know the limits of how much I can push the zoom and crop on a clip without losing any resolution. I have as a practice, been also avoiding zooming in more than 10%, but was hoping to find out a more scientific or mathematical understanding of what goes on when the software zooms in on the footage.

In practice, I have on rare occasions zoomed in up to 100% on some projects for clients and they have not noticed any difference in the resolution – most likely because, like most lay people, they really would not have a trained enough eye to know the difference – but it would be good to know the exact science behind it.

I guess it all boils down to how the software actually manipulates the clip on a pixel level… or maybe even the codec level?

[Blaise Douros]

Gotcha. The best person to answer this would be one of the Adobe guys (are you reading this, Todd Kopriva?), but from what I understand, it basically works something like this (and I apologize if this is overly basic or if you already understand how this works):

When you have a 1920×1080 clip on a 1920×1080 canvas, the pixels are represented on a 1:1 basis. When you scale it up, you of course lose some pixels to cropping, but the ones that remain present are subsampled across the 1920×1080 canvas. Let’s use a single black pixel on a white surface as our clip example: when you increase the size of the clip by 200%, that single black pixel would be subsampled across four canvas pixels (not two, because remember it doubles in both x and y dimensions). Now you have four black pixels on a white field.

This is why resolution loss becomes obvious pretty quickly–when you double the size to 200%, you’re not actually cutting your onscreen pixels in half–you’re cutting them down to a fourth.

Now, let’s say we didn’t scale that image up by 200%, but by only 150% (100% is the native size of the clip). Theoretically, that black pixel is scaled up to the size of two pixels; but it’s not that simple, because the size increases in both x and y dimensions. That means that those four pixels that showed black are going to have to try to represent only part of that two-pixel black dot. So depending on that dot’s position, you might have one central black pixel with the ones around it registering grey as a transition, or maybe four pixels showing dark grey. This process is called interpolation, and there are different algorithms that can be used for different results. Basically, it’s the software trying to average out the values for each pixel, based on what’s around it.

Of course, the other factor that we’re not taking into account is the quality of your initial footage. Super clean, 4:4:4 ProRes is going to look a lot different than highly compressed 4:2:0 DSLR footage, which loses some detail to compression. Highly compressed footage introduces additional artifacts that reduce your resolution in a different way–by decreasing the amount of information in the frame. So uncompressed footage is going to hold up to your scaling a lot better than compressed footage will.

So, I suspect that the answer to your question of “is there a hard limit” is…it depends. Depends on the footage, the subsampling algorithm that the NLE is using, and depends on how much you want to scale it up. Like I said, with AVCHD footage, I can usually get away with 10%.

Don’t you love it when the answer is “it depends?” I sure do 🙂 But knowing about all the factors definitely helps when making an educated guess.

[Devinda Fernando]

Wow, that was a Fantastic explanation (assuming its all true)… Thank you.

***when you increase the size of the clip by 200%, that single black pixel would be subsampled across four canvas pixels (not two, because remember it doubles in both x and y dimensions). Now you have four black pixels on a white field. ***

That also explains the last guy’s point about why the Software’s scaling calibration quadruples the number of Pixels, and not doubles it, when it comes to increases of 100% in size.

*** This process is called interpolation, and there are different algorithms that can be used for different results….

Basically, it’s the software trying to average out the values for each pixel, based on what’s around it. ***

So sticking close to your example… if we had 4 Pixels on a white canvas and those Pixels were 2 x Red and 2 x Orange, then a scale up of 50% would cause the Software to interpolate those existing pixels and create two additional pixels (now total of 6 pixels), that were a cross between the two colors? Yes/no? So in essence, you would be left with 2 x Red, 2 x Orange, and 2 x Red-Orange crossbreed Pixels – am I understanding that correctly?

*** Of course, the other factor that we’re not taking into account is the quality of your initial footage. Super clean, 4:4:4 ProRes is going to look a lot different than highly compressed 4:2:0 DSLR footage, ***

OK, that really cleared up another nagging question I had. But does H264 footage ‘JPEG’ an image? As in, when it compresses an image, does it block/lump similar or same color pixels together? Take for example your example: Instead of a black pixel on a white canvas, lets assume that the pixel is a square of 100×100 black pixels… does the H264 Codec then just create a mathematical representation of that entire square as a single X and Y coordinate and a single Color Space data point for all 10,000 pixels within that square? I guess what I’m trying to understand is how the actual ‘JPEGGING’ works in the compression?

*** Don’t you love it when the answer is “it depends?” I sure do 🙂 But knowing about all the factors definitely helps when making an educated guess. ***

No, No, I Love it… So I think I’ve learned some very important points from this:

So RAW footage (i.e. 4:4:4 ProRes) is actually far more detailed than .H264 compressed footage in terms of Pixels AS WELL AS Color Space, simply because the Pixels in H264 are ‘JPeged’ into groups to save space(Memory). So while the two images have the same Number Of Pixels on each frame, the RAW footage has more detail per pixel, whereas the H264 footage has created blocks of pixels of the same Color Space and thus lost a lot of detail.

So that being said would it be logical to assume that a RAW file, which if we take the example of REDCODE R3D (which is about 160GB per 1 hour of footage for 2K), when compared with H264, which is approx. 16GB for 1 hour of footage for 2K) – the comparable scaling for the R3D would be roughly 10 times that of the H264…? – Because the H264 has been already compressed to a tenth of the size of the RAW footage? So the visible resolution of up-scaling a RAW file to 100% of its original size, would have the same ‘look’ to the human eye, as H264 footage scaled by only 10%. Is that logical, or have I derailed?