Simple question about 4k

[jnorman34]

So, I have a GH4 on preorder and plan to start shooting my architectural documentary footage in 4k. 

 

My question is - If I playback 4k video footage on a 1080 FHD display (like my TV), will I see any difference than if I had shot the same footage in 1080?  If yes, please explain why.

 

Thanks.

[maxotics]

Interesting question, here's my stab at it.  When you shoot 1920x1080 you're really shooting 480x270 red pixels, 480x270 blue pixels and 960x540 green pixels.  These pixels are arranged in a "bayer" pattern.  Before creating an image, the camera or your computer, de-bayers each pixel, a fancy name for taking the missing color information from neighboring pixels (red borrows green blue, etc).  Add to that, the greater number of green pixels are more sensitive to light then the red and green, so you're creating an image with less red and blue than those percentages indicate. 

 

It works pretty well in practice, but in the fine details there are problems where the neighboring color is NOT correct for that pixel.

 

when you shoot 4K and scale down, you are creating pixels (4 into 1) that have no borrowed information because you are blending the three colors into one unique pixel.  Moire and aliasing should be greatly reduced.

[jnorman34]

Danke max

[Guest Ebrahim Saadawi]

Yes of course 4K makes better 1080p. I am excited by the 4K revolution only because we will finally get perfect 1080p and even 720p!

[etidona]

Hi, it depends also on the camera you shoot with. The Eos cinema line for example has a true full hd image because every frame is downscaled from a 4k sensor. In that case i suppose you wouldn'n se any difference. At the least theoreticaly.

[Michael1]

I agree with Maxonics.  I will also add, that if you look at camera tests, every 1080 camera I have seen actually outputs a "measured" resolution somewhat below the 1080 supported by the file type, usually 750 to 850 lines.  If you shoot in 4K, you should easily be able to get the full 1080 after downsampling.

 

Michael

[maxotics]

I will also add, that if you look at camera tests, every 1080 camera I have seen actually outputs a "measured" resolution somewhat below the 1080 supported by the file type, usually 750 to 850 lines.  

 

I was just looking at some 4K footage.  You must be right, it looks a lot better than I thought it would.

[jcs]

Higher end cameras debayer from a higher res Bayer array. The C300 uses a 4K sensor, which averages the G to get 1080 and takes R and B as is: its charts show 1000+ lines res for 1080p. The Sony FS700 does around 1000 lines from a 4k sensor. The 'ancient' consumer Panasonic TM700 does around 1000 lines by using 3 sensors- no debayering. 5D3 1920x1080 RAW with a high quality post debayer looks like 900 lines or so (luminance)- too expensive to compute in-camera.

1000 lines refers to horizontal line pairs before extinction of detail: a measure of horizontal resolution (not related to 1080p vertical resolution). From 1920 samples we can get 960 elements before aliasing starts (Nyquist sampling theory). 1000+ lines is quoted instead of 960 as there is usable detail after aliasing starts (usually cut off at some point with an AA filter).

[maxotics]

From 1920 samples we can get 960 elements before aliasing starts (Nyquist sampling theory). 1000+ lines is quoted instead of 960 as there is usable detail after aliasing starts (usually cut off at some point with an AA filter).

 

Hi jcs, Can you explain that a bit more?  Thanks.  Also, I was always intrigued by the TM700, you make it sound like a very nice camera is acquired cheap.  Have you used one?  Thoughts?

[jonpais]

Hi jcs, Can you explain that a bit more?  Thanks.  Also, I was always intrigued by the TM700, you make it sound like a very nice camera is acquired cheap.  Have you used one?  Thoughts?

I started out with the TM900, which is pretty much the same camera as far as I know, and it's an excellent consumer camcorder. Obviously it's not going to give you the flexibility of something like a GH3, but for recording family events or vacations it's more than acceptable. And then I've seen some shorts shot with the camera as well which looked pretty amazing when I first saw them. The last time I checked, SlashCam rated that line of Panasonic camcorders fairly highly.

[maxotics]

The last time I checked, SlashCam rated that line of Panasonic camcorders fairly highly.

 

Great, thanks jonpais, another site to procrastinate on  ;)

[jcs]

Hi jcs, Can you explain that a bit more?  Thanks.  Also, I was always intrigued by the TM700, you make it sound like a very nice camera is acquired cheap.  Have you used one?  Thoughts?

 

Nyquist-Shannon sampling theory shows us that we need twice as many samples as our target frequency to prevent aliasing. This is true for audio and video and all sampled systems. For example, 48kHz audio can reproduce 24kHz without aliasing. All frequencies above 24kHz must be filtered during recording and playback (in practice higher sampling rates are used to allow for lower cost analog filters). For video we're sampling light instead of sound and the same theory applies. 1920 pixels can represent a max of 1920/2 = 960 'pixels pairs' (on/off) before aliasing. For example, a chart with black and white vertical lines can be used to measure 960 lines pairs from a 1920 image. Even if the camera sensor and chart aren't perfectly aligned, we can see 960 line pairs. When looking at even higher resolution lines, we can see more than 960 line pairs are visible, but with aliasing artifacts (e.g. on the right side of the vertical line chart): http://www.graphics.cornell.edu/~westin/misc/ISO_12233-reschart.pdf

 

Top cameras have optical antialiasing filters which cut off frequencies effectively above the sensor aliasing frequency. These cameras exhibit very high quality images, especially important for moving images/video. While aliasing can provide a false appearance of higher detail, it's one of the give-aways the image/video is not film. If the aliasing is very high frequency, and the image is somewhat noisy, it's not as apparent (e.g. BM cameras with no AA filter). Again, top cameras have very good AA filters (changeable/removable) such as the ARRI Alexa and Sony F55. When looking at a chart we see excellent low-aliasing.

 

The TM700 was the top camcorder (by a quite a bit), IQ-wise in its price range when it was new. 3x 2.53Mpixel sensors (no debayering or related artifacts!), Leica DICOMAR lens (F1.5 - 2.8), 1080p60 support for slowmo, an excellent power OIS, decent audio quality with an external mic (indoors- only neg. issue with camera is fan noise picked up on internal mics). It wasn't very good in low light, however a simple LED light on top of camera, with a shotgun external mic makes a nice ENG/interview camera for any lighting condition. I still have mine but haven't used it much since getting the 5D3. The 17Mbps AVCHD codec while pretty good is a bit over-compressed for large motion scenes/handheld, etc. (60p is 28Mbps). An external recorder might help but I haven't tried it. The zoom range is very impressive: 35mm-420mm (35mm equivalent). With power OIS, even at max zoom, image is pretty stable. Walking while shooting also works very well- no rig needed! :). The autofocus also works pretty well (easier for small sensor cameras).

 

http://shop.panasonic.com/shop/model/HDC-TM700K?t=specs&support#tabs

 

The TM900 is a newer version with slightly better IQ (looks like more contrast):

http://camcorder-test.slashcam.com/compare-what-i-cmp-u-cmd-i-view-u-mode-i-docompare-u-lang-i-en-u-id-i-167-y-185-u-name-i-Panasonic%20HDC-TM900-u-bname-i-Panasonic%20HDC-TM700-u-cmd-i-vergleich.html

Note comments: "Sharpness and color at the AVCHD limits". 

 

Thus the 4K GH4 with one of Panasonic's OIS autofocus lenses should be quite spectacular for doc/ENG/(narrative- some shots).

[nazdar]

 

... Add to that, the greater number of green pixels are more sensitive to light then the red and green, so you're creating an image with less red and blue than those percentages indicate. ...

And I thought (or was reading it long time ago) that our eyes can recognise way more shades of green than other colors therefore they put more green pixels on the chip. True / not true?

[maxotics]

And I thought (or was reading it long time ago) that our eyes can recognise way more shades of green than other colors therefore they put more green pixels on the chip. True / not true?

 Yes, I believe that's true.  Our sensitivity to contrast is much higher than our sense of color, so they use green, which as you say, we're more sensitive to, to capture better contrast than color.  

 

When I was a kid all firetrucks were red.  Then they said we're actually very blind about red in the night, so they painted them yellow (closer to green).  Now they seem back to red.  Guess Firemen would rather look badass in the day ;)

 

The other thing to keep in mind that most of our eye-sight is mostly color blind.  It is ones the cones, which are in the center of our eyes, a few degrees of view, that discern color.  A lot of what we think is color is only our brain filling in the pieces.