GH2 GH3 w/ HDMI External Recorder

Jim Simon said:
In absolute terms maybe, but relative to the rest of the image? (Which is what I'd presume really matters here.)
Click to expand...

Yes in absolute math terms. "4:4:4, 4:2:0" - these subsampling terms are meaningless without stating the accompanying resolution

If we assume Barry is correct about the equivalent deliverable chroma resolution being somewhere between 4:2:0 and 4:2:2, even after debayering and then subsampling to an intermediary 4:2:0 format - which is the worst case scenario and you would never do that - 4K 4:2:0 YCbCr has higher chroma resolution than the best case 1080p 4:4:4 chroma resolution.

If we assume 16:9 4K to be 4096x2304, 4:2:0 subsampling would mean CbCr contain a maximum of 2048x1152 resolution

If we assume 4K , 4:2:2 subsampling, CbCr would contain a maximum of 2048x2304 resolution

Unsubsampled 1920x1080 (or 4:4:4) can only contain a maximum of 1920x1080 chroma resolution

4:2:0 subsampled 1920x1080 contains a maximum of 960x540 chroma resolution
 
Last edited:
All of this to state, as I have said from the beginning, that if you know your way around grading and color correction, you can achieve a lot of what can be found in higher quality chroma outputs, it's just going to take a little more effort working in your secondaries. The chief advantage of an Alexa is the ease/speed with which you can pull an image in any direction you want and the clean quality of the output. However, I am quite confident that even with a DSLR far more can be achieved than most people are aware of, it's just going to take a little more time and elbow grease. The real issue is less the color sampling than the artifacts. Low resolution smearing, aliasing and poor roll off are of far greater concern. A lot of what people refer to as tearing, is actually poor color correction technique. Of course a serious colorist will always prefer to work with great output from the beginning, and many of them hate DSLR's, but if you gave them enough time and a properly lit or exposed shot, they could still probably sort a situation out. When you consider that their job consists of grading shots in seconds you can understand their natural animosity to less than pristine footage, but for most of what we do the need for example of an Alexa is a moot point.
 
Last edited:
PDR said:
Yes in absolute math terms. "4:4:4, 4:2:0" - these subsampling terms are meaningless without stating the accompanying resolution

If we assume Barry is correct about the equivalent deliverable chroma resolution being somewhere between 4:2:0 and 4:2:2, even after debayering and then subsampling to an intermediary 4:2:0 format - which is the worst case scenario and you would never do that - 4K 4:2:0 YCbCr has higher chroma resolution than the best case 1080p 4:4:4 chroma resolution.

If we assume 16:9 4K to be 4096x2304, 4:2:0 subsampling would mean CbCr contain a maximum of 2048x1152 resolution

If we assume 4K , 4:2:2 subsampling, CbCr would contain a maximum of 2048x2304 resolution

Unsubsampled 1920x1080 (or 4:4:4) can only contain a maximum of 1920x1080 chroma resolution

4:2:0 subsampled 1920x1080 contains a maximum of 960x540 chroma resolution
Click to expand...


As I understand it, you have to downsample the pixels in a Bayer pattern by around 25% to 30% to get complete color information for every pixel.

May I therefore conclude that the 4:2:0 format misses only something like 30% of the color information of the 4:4:4 format?

See also Barry's remarks in the quote above.
 
Last edited:
PDR said:
Unsubsampled 1920x1080 (or 4:4:4) can only contain a maximum of 1920x1080 chroma resolution
Click to expand...

True, but that's full color info for every pixel, which seems like it would be better than less than full color for every pixel. Resolution is irrelevant. Percentage of color data per pixel is what counts here. No?
 
Last edited:
Any chance we'll ever see digital sensors where each photosite can detect red, green or blue? If so is there any merit to adding additional colors like orange, purple and yellow detection?

How would any of this affect color information and image processing? Would it be an overall improvement or just a waste of time for a company to pursue sensors like this, if even possible?
 
roxics said:
Any chance we'll ever see digital sensors where each photosite can detect red, green or blue? If so is there any merit to adding additional colors like orange, purple and yellow detection?

How would any of this affect color information and image processing? Would it be an overall improvement or just a waste of time for a company to pursue sensors like this, if even possible?
Click to expand...

Yeah in a 3 chip camera. In a single? It is kind of hard but nothing is impossible. It kind of comes down to the laws of physics. Since sensors detect light values they don't really look at color. A single senor has to "color" each pixel so it can give a color value. You can only really put one color filter on each pixel with the way light works. There was a Foveon sensor which basically did what you want but I haven't heard much about those recently. I believe they are super expensive to create.

http://en.wikipedia.org/wiki/Foveon_X3_sensor
 
Thomas Smet said:
You can only really put one color filter on each pixel with the way light works.
Click to expand...

But the light itself does have color, otherwise even the filters wouldn't work.

Here is another crazy idea, what about a spinning filter wheel with 1/3 being Red, 1/3 blue, 1/3 green. Spin it three times faster than the frame rate over a single monochrome sensor. Would that not work similar to a 3-chip system without the expense or increased flange to focal distance of three chips with a prism?
 
Last edited:
bvidec said:
As I understand it, you have to downsample the pixels in a Bayer pattern by around 25% to 30% to get complete color information for every pixel.
Click to expand...
No, you'd have to downsample by 50% to get complete color information. Which is why the C300, for example, has a 4K chip in order to produce a 2K frame (well, QuadHD chip making an HD frame, but basically the same thing).

The 25-30% number you're referring to is how much resolution is lost when converting from Bayer raw to a video signal, in terms of luminance only -- not color. So a 4K Red camera delivers about 3.2K of luminance resolution (about 20% less than its stated resolution). But it still only has 2k worth of color resolution.

May I therefore conclude that the 4:2:0 format misses only something like 30% of the color information of the 4:4:4 format?
Click to expand...
No. 4:2:0 has 25% of the color information of a 4:4:4 system. 4:4:4 means that every single pixel in the frame has individual red, green, and blue samples. But in 4:2:0, one color sample is applied to every block of 2x2 pixels. One color sample per four pixels, vs. 4:4:4's one color sample per 1 pixel.
 
Jim Simon said:
True, but that's full color info for every pixel, which seems like it would be better than less than full color for every pixel.
Click to expand...
In an ideal world, yes. But, full color for every pixel takes up much more data space than averaging color across multiple pixels does. And when it comes to what the eye can perceive, we're much more attuned to brightness than we are to color. Television has always "cheated" by reducing the amount of color information, because we organic beings really don't see any difference. Higher color sampling is highly beneficial in effects and keying work, but doesn't really make THAT much of a difference in an overall image. Hence why every blu-ray ever made is using 4:2:0 -- it looks "good enough" for people to accept.

Percentage of color data per pixel is what counts here. No?
Click to expand...
Depends on what you're trying to accomplish. If you want the maximum resolution from the fewest amount of pixels, a bayer sensor is the best choice for that. If you want maximum color fidelity from your available pixels, a three-chip or rgb stripe system is more appropriate for that.
 
roxics said:
Any chance we'll ever see digital sensors where each photosite can detect red, green or blue?
Click to expand...
Not likely, but there are such things as the Foveon, which stacks three chips into one and therefore has perfectly co-sited red, green, and blue detection.

Would it be an overall improvement or just a waste of time for a company to pursue sensors like this, if even possible?
Click to expand...
There's reams of debate about Foveon vs. Bayer all over the internet. Foveon proponents say you get better imagery because you have better color and ZERO moire and aliasing issues. Bayer proponents say that you get much higher resolution out of Bayer. There are advocates on both sides, but as per the market, Bayer is used by almost everyone. Foveon is a privately held corp, bought out by Sigma, so only Sigma can use patented Foveon technology.
 
roxics said:
But the light itself does have color, otherwise even the filters wouldn't work.
Click to expand...
Yes, of course -- but you only need r, g, and b to recreate all colors. A sensor can only detect the presence of light, it can't tell what color that light is. Sensors are monochrome. You need to have the filter in place to separate the light into its fundamental r, g, and b components. Otherwise you'd just have a mess.

Here is another crazy idea, what about a spinning filter wheel with 1/3 being Red, 1/3 blue, 1/3 green. Spin it three times faster than the frame rate over a single monochrome sensor.
Click to expand...
Sounds like a DLP. It works, but some people are susceptible to seeing a "rainbow" effect because of it. Actually the newest crop of LCoS viewfinders in the Panasonics use similar technology, and the viewfinder in the GH2 is an example. Flick your eye left or right quickly, and you'll see a red or green or blue "afterimage" briefly displayed. It's okay for some, but an annoying artifact for others.
 
Jim Simon said:
It does, thanks.

Sounds like RAW using three full resolution CCDs is still the best image possible outside of real film.
Click to expand...

Yeah, 3-chip is far from dead yet, especially for greenscreen, where stuff like chroma resolution matters. On the other hand, as chip resolution increases, this advantage of 3-chip over Bayer hits a point of diminishing returns. After all, if you're shooting at something crazy like 8K, does it really matter if you have "only" 4K worth of chroma?

I'm still waiting on my perfect camera. 3 medium-format 8K chips (don't need 16K just yet), 32bit raw, global shutter, 24 stops dynamic range. Price: $10 with 4 proofs-of-purchase from boxes of corn flakes. (Actual, real world price on introduction would probably be closer to 2 winning Powerball tickets.) A guy can dream.
 
Jim Simon said:
True, but that's full color info for every pixel, which seems like it would be better than less than full color for every pixel. Resolution is irrelevant. Percentage of color data per pixel is what counts here. No?
Click to expand...

Both luma and chroma contain more resolution from the RED footage, even if we assume Barry's worst case numbers.

You can downsample the Red footage to 4:4:4 1920x1080. That is "full color" at 1920x1080 resolution. Oversampling has great benefits . Going down is easier than going up.

Subsampling terms are meaningless without stating the resolution to which they are referring to. It's like saying something like "80%." Well 80% of a big number is a big number. But 80% of a small number is still a small number.
 
....which translates to us downsampling a GH-1/2/3's image to a 4:4:4 image but only in standard definition. Oversampling is the shiznit!!!!
 
Hmmm wow this is really interesting. I always assumed that raw was the equivalent of 4:4:4 but obviously not. Which actually makes a lot of sense. Recently I was keying some RED footage and I was surprised at how poorly it keyed.

@Thomas Smet mind sharing your technique to get 4:4:4 using filters out 4:2:0 that seems kinda like magic to me. So yeah, I don't believe you, but I hope to god you are right. I will have to say I have gotten similar results to keying SLR footage than from poorly handled red footage. (ie RED footage transcoded to prores 422 1080)
 
hmhmh..

Isn't GH3's HDMI ALWAYS 1080p60 ( 1080p50 in PAL ) ??? So, it is impossible to get cheap recorder for 1k$ ...

.
 
You must log in or register to reply here.
Back
Top