Do codecs make a difference in image quality?

[Ty Ford]

There are 5 different codecs in the HM650. The new 50 Mbps H.264 doesn't seem to make a lot of difference over the other 35 Mbps codecs. Does anyone have any info on which codecs look better?

Thanks,

Ty Ford

[Peter Manojlovic]

Codecs are all subjective...
But your question isn't about codecs, as much as bitrate for the same codec..
Nonetheless, both bitrates should be adequate to handle your image (assuming it's 1920x1080).

[Andrew Kimery]

All other things being equal the higher bit rate should look better. If you 'stress' the codec out (a tight shot of flames in a fire place, a leafy tree being blown in the wind, flash photography, agressive grading/manipulation in post, etc.,) then you'll probably start seeing differences between the 50Mbps version and the 35Mbps version.

For a poor analogy, it's like having one car than can drive smoothly up to 50mph and another car that can drive smoothly up to 35mph. If you are only driving at 25mph then either car will work but if you are going 45mph then you'd want to take the car that can drive smoothly up to 50mph.

[William Hohauser]

They should make a difference in theory but in practice that would be up to the capabilities of the equipment. The differences between h264 and MPEG2 will also create a subtle visual change. 50Mbps should look the best with 24p looking better than 60p in individual frame clarity although the frame rate in 24p might reduce action clarity so there is a trade-off.

[Bruce Watson]

Originally Posted by Ty Ford

There are 5 different codecs in the HM650. The new 50 Mbps H.264 doesn't seem to make a lot of difference over the other 35 Mbps codecs. Does anyone have any info on which codecs look better?

Look better, for what kinds of work? If your work involves lots of motion, as in lots of camera moves so that everything in the image changes with every frame recorded, then Intraframe CODECs tend to do better because they compress each frame individually. That is, they lessen the motion artifacts. These CODECs tend to run high bit-rates because they are compressing the entire frame, every frame, which takes a lot of bits to do.

If your work involves lots of fixed camera positions without camera moves, say a musician sitting in a chair on a stage, then an interframe CODEC typically does fine. It can do this because it's a LongGOP (long group of pictures) CODEC which compresses the first frame fully like an intraframe CODEC, then for the rest of the frames (pictures) in the group, it subtracts the current frame from the initial frame (I-frame) to get the differences, and just compresses the differences. Clearly, this is a much higher amount of compression, and results in a considerably lower bit-rate. But it also takes considerable NLE computer horsepower to play it back because of all the math involved in rebuilding each frame.

If your work involves lots of chroma-key work (green screen, blue screen), you'll usually get better results from a CODEC that supports at least 4:2:2 chroma subsampling. Most consumer level CODECs are 4:2:0, and therefore don't record much of the chroma data, which of course makes chroma-key work problematic if your NLE can't correctly and reliably identify the key color. Professional cameras often use CODECs that record 4:4:4 (which is all of the chroma data available), but this of course takes more bits, and therefore a higher bit-rate.

If you work involves a lot of color grading, you'll usually get better results from a CODEC that records at least 10 bit 4:2:2. More bits is better, as is more chroma data. So many professional cameras use CODECs that record 12 bit 4:4:4. Which takes more bits and therefore a higher bit-rate. There's a pattern here...

So... what are you trying to capture? The easier it is to capture, the less difference you'll see between CODECs.

[Ty Ford]

So, more high frequency video information (leaves, water surface), higher bit rate?

Other than that, all the 35s look pretty much the same? No need to experiment with them to see if one does better than the others?

I'm just looking for ways to increase definition and reduce noise and crushing of the blacks.

Thanks for all of your comments. All points well taken.

Regards,

Ty Ford

[Bruce Watson]

Originally Posted by Ty Ford

So, more high frequency video information (leaves, water surface), higher bit rate?

It's not that simple. It's not that you need a higher bit-rate so much as you need a CODEC better suited for what you're trying to accomplish. For a frame that's nothing but moving water, an intraframe CODEC will likely perform better than a LongGOP CODEC. Not that it will for certain, because there's more to it than just the CODEC.

Originally Posted by Ty Ford

Other than that, all the 35s look pretty much the same? No need to experiment with them to see if one does better than the others?

No such luck. You will have to experiment and see what works best in your situation. Sorry.

Originally Posted by Ty Ford

I'm just looking for ways to increase definition and reduce noise and crushing of the blacks.

That's going to be more a function of the imager, and the image processing in the camera. Some of this is how they implement a given CODEC, but more is how they process the information coming from the imager before it gets passed to the CODEC. For example, one way to increase definition is to over sample a single imaging chip, or to use three chips (RGB). And of course the glass being used clearly influences resolution and contrast of the image.

Similarly, noise and black levels are not really CODEC problems, but are more camera problems. And they may well be due to camera settings not appropriate for the scene being filmed, not using a waveform monitor to check for proper exposure and lighting, etc. Of course, it could be that you're just not getting enough light on the subject. Starving a sensor often shows up as a noisy image that lacks color and tonality.

Hard to say more without seeing a sample. Post a frame if you want; there are people here with a lot of video experience that may be able to point you in the right direction.

[Ty Ford]

The camera is an HM 650.

Here's an in depth technical appraisal.

https://tech.ebu.ch/docs/tech/tech3335_s08.pdf

I'll post some footage soon.

Regards,

Ty Ford

[Ty Ford]

OK, here's footage from this morning.

I'm seeing blacks that don't crawl (without noise), or don't crawl much.

How does this look to you? Is it sharp enough?

https://www.dropbox.com/s/qb7gycizlc...posureTest.mov

Regards,

Ty Ford

[David Heath]

Originally Posted by Bruce Watson

If your work involves lots of motion, as in lots of camera moves so that everything in the image changes with every frame recorded, then Intraframe CODECs tend to do better because they compress each frame individually. That is, they lessen the motion artifacts. These CODECs tend to run high bit-rates because they are compressing the entire frame, every frame, which takes a lot of bits to do.

Practically, it's the last sentence that is most important, the one about bitrate.

A lot of general comments about motion artifacts really only refer to situations where bandwidth is very limited - to the extent that with all else equal an I-frame only codec would look bad ALL the time!

Compare an interframe and an intraframe codec with all else equal (esp bitrate) and the interframe codec will ALWAYS be better than intraframe, albeit needing more processing power to deal with. How much better is difficult to quantify, and all interframe codecs are far from equal. The better ones tend to dynamically shift datarate from I-frames to difference frames according to motion - static sequence, lots of bits for the I-frames, high motion, lots for the difference frames.

In an extreme motion case (every frame different!!) it may allocate such that difference frames get as much as I-frames - which is equivalent to it's becoming an I-frame only codec!!

For professional codecs it's really only when you start to process the video, including decompression and recompressing in a different codec, that the issues start to show up - there's little practical point in looking for differences in first generation video. Even AVC-HD looks pretty good in first-generation - the problems start to show the more post manipulation gets applied.

That's why broadcasters insist on minimum codec requirements - it's nothing to do with how the images look out of the camera, all to do with how they will stand up to the broadcast chain. Best analogy I've heard is with food poisoning (!) - the food can taste great at time of eating, but can make you feel sick later. Same with using a non-broadcast codec - the original pictures can look great, but hidden issues due to compression can cause big problems by the time of transmission

[Bruce Watson]

Originally Posted by David Heath

In an extreme motion case (every frame different!!) it may allocate such that difference frames get as much as I-frames - which is equivalent to it's becoming an I-frame only codec!!

True, but at a low maximum bit-rate. Which is why, for the cases where every frame is different, the intraframe CODECs are better -- but you pay for the decrease in motion artifacts with considerably higher bit-rates and therefore more usage of storage.

[Bruce Watson]

Originally Posted by Ty Ford

OK, here's footage from this morning.

I'm seeing blacks that don't crawl (without noise), or don't crawl much.

I'm not seeing anything that's immediately objectionable.

Shooting in direct sunlight is always a dynamic range challenge for a camera. Unless you're shooting with a Red or Arri, your camera can probably see less than 14 stops from featureless white to featureless black. That is often not quite enough to handle the full range from bright sunlight to dark shadows. In that case, the camera has to shoehorn the existing physical dynamic range into what it can process; it does what it's designers tell it to do. Depending on its design, it might let the shadow detail go which you and I see as crushed blacks. It might let both the top and bottom go, which we see as blown highlights and crushed blacks both. If you handle exposure manually, you have to make that same choice -- something has to be sacrificed.

I'm not seeing much noise here indicating that the sensor isn't likely being light starved. It's probably just a dynamic range limitation.

If you want to see more shadow detail, you'll have to decrease the dynamic range of the scene somehow so that it fits within the range that the camera can handle. The "normal" way is to fill the shadows. Outdoors this is often done with a bounce card. In studios it's often done with a fill light. You can approach it from the opposite side too. That is, tame the highlights. Outdoors this is often done by flying some diffusion between the sun and the subject, or flying a net. In the studio it's usually done by position of the key light, and the use of scrims to block some of the light leaving the instrument, using a dimmer if your instrument can be dimmed, or putting some diffusion between the instrument and the subject (for example, a softbox). I'm just sayin' that there are many paths to the waterfall.

[David Heath]

Originally Posted by Bruce Watson

True, but at a low maximum bit-rate. Which is why, for the cases where every frame is different, the intraframe CODECs are better -- but you pay for the decrease in motion artifacts with considerably higher bit-rates and therefore more usage of storage.

In my (extreme hypothetical) case of every frame being different - the ultimate amount of "motion"! - then with dynamic allocation, and difference frames getting the same allocation as I-frames, the interframe codec effectively becomes intraframe! (And whilst you don't get any motion artifacts, the static artifacting is worse than with less motion - but only the same as an equivalent I-frame only codec.)

To take a practical example of (say) HDV versus DVCProHD, then I'd certainly agree that with high motion levels, DVCProHD will show less motion artifacting - but that's not really inter versus intra that leads to that - far more that DVCProHD has 4x the bandwidth. If we level the playing field and talk of a 100Mbs MPEG2 interframe codec, then such will outperform DVCProHD in virtually any circumstance. Practically, XDCAM 422 at 50Mbs is also generally considered better than DVCProHD.

Also worth noting is that HDV does NOT seem to dynamically allocate bitrate, which is half the problem. That's why 35Mbs XDCAM (which does dynamic allocation) makes such a dramatic improvement, in spite of the same basic technology (MPEG2), and in spite of the bitrate not being that much higher than HDV.

Which is why XDCAM suffers far less from motion artifacting than HDV. With high motion it may be true that each frame gets compressed more in the static sense - but with high motion the blurring of the frame means static artifacting is far less visible anyway.

The exception may be such as ripples on water, or blades of grass when the static artifacting due to all the motion may get seen. But even in the most extreme case, it's never going to worse than an intraframe codec OF THE SAME BITRATE and will normally be far better.

[Ty Ford]

Thank you everyone!

Bruce. Right it was a better day than I expected with that clip. I do understand "usable dynamic range" based on the capabilities of the hardware.

Your explanations are very helpful to me as I continue to wrap my head around what has been elementary science to you for some time.

Regards,

Ty Ford

[Bruce Watson]

David, I think we're in violent agreement.