Resolution Vs Compression

[geminigod]

This is an interesting technical discussion to have that is very relevant to the emerging world of HD fanediting. I wish I had a good evidential answer to share, but I don't. So instead I will discuss the theory in a couple sentences and leave it open for debate.

Resolution and compression are two sides of the same coin when maintaining a constant bitrate. You can never get anything for nothing. Thus, it is a fact that 1080p video will be more compressed than 720p at the same bitrate. I think most would agree that at a high enough bitrate, you get a much greater return on having a high resolution that is slightly more compressed. Agreement to this indirectly implies that at some low bitrate the opposite must be true. If you buy into this premise, then that means that there is a threashold we could theoretically define whereby using a higher bitrate dictates we should encode at 1080p and a lower bitrate dictates we should encode at 720p in order to maximize visual quality.

Do you agree with this threshold idea and if so, what do you think this threshold is? I would guess that it lies somewhere 3-5 Mbps.

 

[Captain Khajiit]

it is a fact that 1080p video will be more compressed than 720p at the same bitrate.

Yes.

Do you agree with this threshold idea?

No. How good a source looks at a given resolution is not simply a function of bitrate. Some sources compress better than others. At a given bitrate, some videos will look good, whereas others will look bad.

There is no way of telling how well a movie will compress in advance. You have to encode it, look at it and make a decision about how it looks for yourself. Any decision you make will of course be highly subjective. While you might be able to make some rules of thumb that help you, you will find that others disagree with these, so they are only really useful for providing you with a place to start when you begin experimenting with encoding a particular movie.

 

[geminigod]

Good point. In comparing the same source material at different bitrates and resolutions, I think we could probably arrive at a threshold number that a group could for the most part objectively agree on, but you make a great point that that doesn't necessarily mean the results would be the same with different source material.

That said, as a general guideline, I am starting to think that somewhere right around 4 Mbps is really the min. for 1080p. I encoded a piece of 1080p video at 4 and it still looked mostly good, but I could see some obvious compression artifact that didn't exist at 4.5-5 mbps. Conversely, I watched Boon's Braveheart last night and it looked pretty clean at 720p with average bitrate of 3.3mbps.

 

[Brumous]

Also I think it makes a difference which encoder you use, how many passes it does, etc.

 

[Captain Khajiit]

^^If you are encoding to MKV with x264, a better approach would be to use constant rate factor (CRF). You set the quality that you want and the encoder uses the bitrate required to achieve that quality, so you don't have to devise a notional bitrate threshold (which is misconceived, in my view). The general consensus on Doom9 and Videohelp is that a CRF of 18 is sufficient to achieve good quality, whereas 22 gives a reasonable result for a small file size: that's why these values appear in presets for GUIs such as Ripbot and XVID4PSP. You can set values in between, of course.

The best thing to do is to encode a sample with a CRF of 18. Then encode a series of samples, reducing the CRF by one each time, until you think you see an appreciable hit in quality. When you do, you've found the level of compression that you find acceptable. Most people tend to stick to this number for their encodings, unless they find a reason to adjust it for a particular encoding; output too big, unexpectedly low quality etc. You will find that people tend to favor a range of values between about 16 and 23 and a range of speed and quality presets, including deblocking values. It really is subjective.

The other advantage to CRF is that you only need to do one pass. A second pass is waste of time when encoding an MKV with x264: it's only necessary when you need to hit a particular file size. You probably already know that MKVs routinely have borders cropped to improve quality and reduce file size.

 

[geminigod]

I'm glad you mentioned that. You are totally right. From a practical standpoint, unless you are trying to target a specific file size, using CRF is the way to go. I have produced very good results much faster that way in the past. But since much of the work done on this site is targeted toward burning to disc where 2-pass bitrate can be useful to maximize the usage of that space, I thought perhaps a little mental masturbation on the subject might be fun.

It is a bit more of a relevant issue for larger-sized projects. Fitting a basic 2-hour movie at a good bitrate is pretty easy. But with my edit pushing almost 3 hours, plus 25 minutes of bonus material, plus commentary, plus menu. These considerations were very much on my mind the past couple weeks while encoding. Both the DVD and AVCHD versions came in at almost exactly 100% disc capacity in size.

 

[Captain Khajiit]

Yes; two passes are required to fit a movie onto a disc. My advice for AVCHDs is to always encode at 720p onto a DVD-9. If you are encoding to a disc, an average bitrate calculator has to be used anyway, and you more or less have to live with the result, unless you use multiple discs -- in which case you might have to question whether an AVCHD release is worthwhile.

 

[geminigod]

Multiple discs definitely not an option!! :-o

Now we are getting to the heart of the matter I was trying to get at! You say always encode at 720p onto a DVD-9. I say resolution can go a long ways in overcoming visible compression. I'm not sure which is best, and I agree it is highly dependent on the source material. What I do know is I did my Matrix AVCHD at 1080p and it looks pretty good to me. Maybe at some point here I will find a piece of it to render at 720p and try comparing.

 

[Brumous]

How about upscaling capabilities in individual users' video players? Will that make a difference in whether resolution or compression produces the better image quality? And the characteristics of different HD monitors, TVs and projectors?

 

[Captain Khajiit]

You say always encode at 720p onto a DVD-9. I say resolution can go a long ways in overcoming visible compression.

By visible compression, I take it you mean macroblocking. Try playing with deblocking too. The absence of blocking is not the only ingredient in making 1080p look good though.

I really don't like the dichotomy you are making between compression and resolution. What I think you are driving at is that grain can be harder to process in a video that has been scaled down, so it often results in blocking, so you are wondering if there might be more obvious blocking in a downscaled 720p encoding with a reasonably high bitrate than a 1080p encoding with a low bitrate. It's possible, but overall image quality will be lower in the 1080p encoding, especially during movement, so I'm not sure you are gaining much by keeping the resolution.

What I do know is I did my Matrix AVCHD at 1080p and it looks pretty good to me.

I've no doubt it did to you, but other people looking at it might not agree. This is one problem with your bitrate threshold idea.

At the bitrate that your AVCHD must have had, 1080p usually looks sort of "flat" and lacks detail to me. I have rarely seen the sort of picture quality one would expect from 1080p at anything less than 10000kbps. It's not impossible, but it's unlikely unless you have a source that compresses extremely well. That's why I recommend that 720p be put on a DVD-9. You see this in "scene" releases, which tend to be about the size of a DVD-9 (DVD-5-size video stream + DTS core) for 720p and 15GB-ish (total) for 1080p. This is with the borders cropped too, mind you, and more reframes and b-frames than are in spec for AVCHD.

 

[geminigod]

We do at least agree that 15'ish is the minimum ideal for 1080p. Below that it becomes a function what your objectives are and what you define as acceptable loss.

One of these days I will sit down and try to conduct an experiment that will produce objective results on the subject. I wonder if a program exists that can analyze two images and compare pixel differences? It would be interesting to create a series of clips, then compare them with a computer and also a blind study with a handful of video snobs rating the various clips. I understand the science behind all this, but it would be nice to see some hard data that better details the scope of influence these various factors can have on the final product.

 

[Captain Khajiit]

We do at least agree that 15'ish is the minimum ideal for 1080p. Below that it becomes a function what your objectives are and what you define as acceptable loss.

I'm still not sure we are seeing eye-to-eye on this issue, but there you go lol.

One of these days I will sit down and try to conduct an experiment that will produce objective results on the subject.

I don't doubt for a second that you understand the science behind this, but I don't believe that the results will ever be anything other than subjective. There have been years of argument over at Doom9 and Videohelp about the issue of quality. You can't even ask about what is best under Doom9 forum rules because it creates so many arguments.

For example, some people don't mind a loss in detail and a soft picture, as long as blocking is kept to an absolute minimum. Others can take some blocking as long as the picture is sharp and detailed. I once encoded a project for a friend and showed him three samples. What I thought was the best he thought was the worse, and vice versa. I could not believe that he picked the encoding that he did, because to my eyes it looked awful, but that's how it was.

 

[geminigod]

Interesting. I guess we'll leave it at that for now! The great debate yet unresolved.