Broadcast quality compression

[BroadcastTVS]

I captured this broadcast tv video sample and wondering how I can get this similar quality out of my video compression, especially from graphic titles. The title at 8 seconds looks "softer" and blends in with the video. However, my titles over my video "sticks out" and looks little sharper (video is same resolution as broadcast sample, 720X576). I may being to picky but my graphics and video looks amateur and not what Im looking for. If you need examples and any more information, please let me know. Thank you. (Does this issue I'm having, have anything to do with this: http://blogs.adobe.com/aftereffects/2013/06/bicubic-resampling-for-improved... )

http://reels.creativecow.net/film/broadcast-tv-image-quality

[LigH.de]

There are indeed several factors which make a flash appear annoyingly obvious. Besides sharpness, which may be reduced by a Gaussian blur with small radius, it is also often the saturation and contrast which exceeds broadcasting limits (you may remember "PC vs. TV range").

[BroadcastTVS]

Here is an example of my video and graphic mock. It's looks "lousy" and lacks the quality to the original video.

My sample: https://reels.creativecow.net/film/poor-quality

Original: https://reels.creativecow.net/film/broadcast-tv-image-quality

You can see the difference. Pay attention more to the title, that is my main issue.

Video settings:

720X576 MPEG2
Encoded from AME

[poisondeathray]

Don't mix "compression" with "graphic design" , those are different concepts

From a technical standpoint, the "original" is worse. There are oversharpening halos, color bleeing, more compression artifacts. (But the graphic has been through 1 more generation loss than your version).


In terms of motion characteristics, there is "pre-ghosting" in the original. This was probably not intended and a side effect of compression


The other difference is subpixel interpolation . The original graphic wasn't placed on full pixel boundaries, but on 0.5. This generates a softer edge. This example shows the difference on the vertical . Vertical softness is more important for broadcast because of interlaced displays, and sharp edges can cause line twittering. Blurring the graphic can achieve a similar effect

full pixel boundary


sub pixel boundary

[pandy]

Do all processing in higher resolution than target, at the end of chain use decent resizer (down). This will give you subsample precision and should remove aliasing.

[Cornucopia]

That only works well if your original background video is ALREADY at a higher resolution than will be finally presented. Otherwise, you risk quality loss through unnecessary resizing (UP, then DOWN again). Makes much more sense to properly prep the graphic.

Scott

[BroadcastTVS]

Thank you posiondeathray, I have much of a better understanding now. Spot on with 'Vertical softness.'

"From a technical standpoint, the "original" is worse"


For me personally the original looks better, and the text looks somewhat "clearer" then the mocked sample. To me, its almost as if I have scaled up this graphic if you look at the edging of the text. I know I may focusing on issues on the graphic, but the outcome on the quality of the graphic has to do with how its compressed, right?

Thank you.

[poisondeathray]

Thank you posiondeathray, I have much of a better understanding now. Spot on with 'Vertical softness.'

"From a technical standpoint, the "original" is worse"


For me personally the original looks better, and the text looks somewhat "clearer" then the mocked sample. To me, its almost as if I have scaled up this graphic if you look at the edging of the text. I know I may focusing on issues on the graphic, but the outcome on the quality of the graphic has to do with how its compressed, right?

Thank you.

You are contradicting yourself now. On one hand you say the original "blends" in more with the video, is softer . Now you say original is "clearer" . You can't have both

It's more "clear" because of the halo artifacts. Those are micro contrast edge enhancements from sharpening, made worse by compression artifacts. Zoom in to picture to see them. Oversharpening halos are "unprofessional"

1:1 crop


4x nearest neighbor enlargement



Certainly the viewing audience doesn't want distracting station logos and overlays. Maybe the station does.

Choose one - either unobtrusive and soft, blending in. Or sharp , delineated edges . You can't have both

So again, from a technical standpoint, the original is "worse"

the outcome on the quality of the graphic has to do with how its compressed, right?

Yes, very much so . Lossy compression throws away bits of data each time. Think of how you save and resave a jpg image over and over again. The "original" here has been through 1 more round of compression compared to yours

[BroadcastTVS]

You are contradicting yourself now. On one hand you say the original "blends" in more with the video, is softer . Now you say original is "clearer" . You can't have both

I certainly am. Now that I'm aware of the difference between full pixel boundary and sub pixel boundary I can understand why the graphic looks softer and "blends in" and thats one issue sorted. Also, how did you achieve the sub pixelation boundary in your example?

from a technical standpoint, the original is "worse"

I don't believe so IMO, due to the fact this was captured from a broadcast TV station and what you see in the example is what you'll see when viewing, it has to be somewhat professional?. I may be going off-topic and appreciated the help

[poisondeathray]

You are contradicting yourself now. On one hand you say the original "blends" in more with the video, is softer . Now you say original is "clearer" . You can't have both

I certainly am. Now that I'm aware of the difference between full pixel boundary and sub pixel boundary I can understand why the graphic looks softer and "blends in" and thats one issue sorted. Also, how did you achieve the sub pixelation boundary in your example?

I used After Effects. You mentioned you did the overlays/graphics in AE . Just position the graphic to 0.5 instead of integer values. e.g. if original Y position was 100, move it to 100.5, or 99.5. Or vertical blur would achieve similar effect

from a technical standpoint, the original is "worse"

I don't believe so IMO, due to the fact this was captured from a broadcast TV station and what you see in the example is what you'll see when viewing, it has to be somewhat professional?. I may be going off-topic and appreciated the help

I'll re-iterate - on a technical level, the "original" that you presented is clearly worse because it has halos, compression artifacts, worse chroma bleeding . Those are "technical" aspects when you look at video. Do you think the original broadcast master had compression artifacts? Do you think they degraded it on purpose ? Absolutely not. Not a chance in hell. The sharpening / halos are a combination of both , post sharpening and it's enhanced by the compression artifacts

Do you understand that this isn't really a valid comparison ? (It's not really an "apples to apples" comparison) . You're comparing different generations. So you can degrade it, you can blur it, you can oversharpen it, you can add artifacts if you want , but you should understand everything goes "downhill" . Quality starts at the source. Everything gets worse after each stage. If you degrade it at the beginning, it's only going to look worse when it's actually broadcasted, when viewer actually sees it


Added some blur, adjusted the levels, sharpen it a bit .

[BroadcastTVS]

Thank you, very informative and have gained a lot more knowledge from that.

[pandy]

That only works well if your original background video is ALREADY at a higher resolution than will be finally presented. Otherwise, you risk quality loss through unnecessary resizing (UP, then DOWN again). Makes much more sense to properly prep the graphic.

Scott

Nope - point resampling is perfect from mathematical perspective, also resizing with edge detection provide subjectively good results (but this is unrealistic slow to be used in real flow) - point resampling of source will not introduce any artifacts.

( from mathematical perspective point resampling with even factor is perfect http://en.wikipedia.org/wiki/Zero-order_hold )

[jagabo]

point resampling is perfect from mathematical perspective

But it looks like shit.

[ndjamena]

point resampling is perfect from mathematical perspective

But it looks like shit.

...but not if you intend to resize it back down to it's original resolution later, surely?..

[jagabo]

point resampling is perfect from mathematical perspective

But it looks like shit.

...but not if you intend to resize it back down to it's original resolution later, surely?..

Yes, if you downsize again with the same point resize (ie, no rounding differences) you'll end up with exactly what you started with. But it makes no sense to do this in the context of overlaying with quarter pixel (or finer) accuracy. It's the same as overlaying at the original size.

[pandy]

point resampling is perfect from mathematical perspective

But it looks like shit.

...but not if you intend to resize it back down to it's original resolution later, surely?..

Yes, if you downsize again with the same point resize (ie, no rounding differences) you'll end up with exactly what you started with. But it makes no sense to do this in the context of overlaying with quarter pixel (or finer) accuracy. It's the same as overlaying at the original size.

Upsampling with even point resample, proccessing, down resampling with spline - still looks like shit?

[budwzr]

And moving on to the design element, that banner is like something like from the Archie Bunker era. The channel logo looks passable, but that banner needs an updesign.