"Fringes" around objects and text

[rbruner2]

I'm trying to convert a movie I made (720x480 AVI) to VCD format using TMPGenc. The problem I keep having, and have yet to find any fix for is when there is a still image either in the forground or background. For example, if I have title text floating on top of a motion background, or if there is a person talking with a stationary background.

Anyhow, around the edges of the stationary object (or moving object, depending on how you look at it), there is this blocky fringing. I haven't found any way to get rid of it. I'm seen it both in SVCD and VCD conversions. (Currently, I'm trying to do a straight VCD conversion because I'm going to distrubute this to friends/family, and I need it to be the most compatible possible.)

Any suggestions on how to reduce or eliminate this?

[SaturnBlader]

Sometimes it depends on how you capture. I would recoment capturing using the DivX 4.12 Codec with a high bitrate setting or if you have a lot of space on your HDD (20GB for 1 Hour 45 Min captures) the use the DV Codec.

[kinneera]

Actually, what you're seeing is known as Gibb's artifacts, if I recall correctly. It is a phenomenon that occurs when you attempt to encode hard edges at low bitrates. You can reduce the effect, but due to the technicalities of MPEG encoding, the tradeoff is that you will get more blockiness in solid low-complexity areas.

[rbruner2]

SaturnBlader:

Capture? Part of it is CG animation, some of it is DV footage captured via Firewire, and the rest is editing I did using Studio 7 (on-screen titles).

kinneera:

You said I can reduce the effect...but how? Bit rate is fixed for VCD (I don't want to create a non-standard VCD). So, what else needs to change?

[kinneera]

Well, increasing the bitrate is the best option (XVCD, SVCD), unfortunately this isn't an option with straight VCD.

More specifically I was referring to the "quality" slider in CCE, where lower numbers focus on reducing Gibb's artifacts, whereas high numbers do the reverse. Unfortunately, CCE is a (self-admitted, even) poor encoder for VCD, and I don't know of any similar control in TMPGEnc.

What you can do is apply some degree of spatial smoothing (noise reduction filter in TMPGEnc) to soften hard edges and make it a little easier on the encoder. Only you can decide the level of tradeoff that is acceptable in terms of softening vs artifacts, though.

Also, you can take heart in the fact that Gibb's artifacts tend to be fairly well hidden when viewed on a television screen, even if they seem painfully obvious on your computer screen.

[rbruner2]

Also, you can take heart in the fact that Gibb's artifacts tend to be fairly well hidden when viewed on a television screen

Actually, when I viewed my movie on the TV, the artifacts became painfully obvious to the point of distracting, whereas on the computer screen, they are barely noticable. (However, other things are less obvious on the TV...but I find that, in general, the TV makes flaws much more apparent than on the computer screen. I don't know how many times I've encoded something, previewed it on my computer, and thought I was happy with it until I played it on my TV.

What you can do is apply some degree of spatial smoothing (noise reduction filter in TMPGEnc) to soften hard edges

Actually, noise reduction doesn't soften hard edges at all. What it does is look for "artifacts" that show up in one frame but not in the next. Basically, anything that shows up for a single frame is considered noise. These artifacts, however, are not reduced with the noise filter...I already tried that one. There are a bunch of filters, though, so I'm sure SOMETHING must help. I just can't believe others haven't had this problem. EVERY movie I've converted has had them so far.

I've never used CCE, nor will I given the price. So, I'm looking for a TMPGenc solution.

[kinneera]

Odd, I'm guessing you have a large and/or high definition television? For the majority of individuals, such artifacts are reduced, not exacerbated when viewed on their television, although there are no guarantees as obviously your experience has been otherwise.

As for noise reduction, you are absolutely right, but only halfway (at least in TMPGEnc). There are two types of noise reduction - temporal (the one you described, which compares across frames) and spatial, which looks within the frame for pixels that do not appear to match the surrounding area with reasonable accuracy (tolerance usually specified by user) and blurs them together. The noise reduction filter in TMPGEnc appears to provide both in one easy-to-use package. The "still picture" slider adjusts the spatial smoothing, and the "time axis" slider adjusts temporal smoothing. The range option (appears) to be roughly analogous to the radius option in a typical spatial smoother.

Basically what this means is that a high "still picture" value with a high radius will provide a smoothing effect, which may produce better results for you.

Using the edge sharpening filter with a negative value may also help.

[rbruner2]

what this means is that a high "still picture" value with a high radius will provide a smoothing effect, which may produce better results for you.

I played around this this last night, and it helped minimally. (I didn't even bother checking it on the TV, since I could still see the artifacts on my computer.)

I then tried the Advanced option..I forget what it is called now...but it is at the bottom of the Quantize Matrix pane. That didn't do much either.

Then I tried just putting everything back to normal, except I upped the bitrate to 1800 instead of the normal ~1100. This actually helped some, though not completely. I suppose if I upped that to 2300, it might help enough to make it not noticable. But, again, then I'm working with a non-standard VCD. If I use 2300 bitrate for SVCD, the resolution has to be 480x480, which, I presume, means that the bitrate won't be high enough to avoid the problem. (That is probably why when I generate SVCD's, I get the same problem, despite the higher bitrate.)

To me, this problem makes tmpgenc of little value. I've yet to generate ANY movie without these artifacts. If tmpgenc is supposed to be one of the best in quality, does that mean EVERY encoder experiences these artifacts???

One thing I might try to converting the movie into a DivX format first, and THEN convert to VCD MPEG using Tmpgenc. I did this one other time, and I ended up with a slightly better quality video...although, I don't know about these artifacts, since when I ran that test, there were no stationary objects.

Any other suggestions from anyone on how to avoid of minimize this?

[gonz0]

VirtualDub has a number of filter options (some default ones and more that you can download). You could apply filters in VirtualDub and then frame serve to TMPG, but I don't know if any of these filters will fix your problem. Some like to do the resizing vdub as well.

Otherwise, I assume you've tried these settings in TMPG.
Set the Motion search precision to high or highest
Quantize matrix options:
- try different quantize matrices
- soften block noise
- Use floating point DCT
- enable or disable "no motion search for still picture"

[cuervo]

not sure about other codec, but, anything encoded/captured with a DV codec(this includes ALL DV video cameras) will exhibit artifacts called "mosquito noise". Mosquito noise is a "halo" surrounding areas of high contrast, like dark horizon and light sky. Unfortunately, mosquito noise is the result of the compression algorithm used in DV. Nothing you can do about it, but, in my experience, it's not noticeable on a video.

[rbruner2]

gonz0:

Otherwise, I assume you've tried these settings in TMPG.
Set the Motion search precision to high or highest
Quantize matrix options:
- try different quantize matrices
- soften block noise
- Use floating point DCT
- enable or disable "no motion search for still picture"

I tried high/highest motion precision. No change.

I tried soften block noise up to values of "50" for both non- and intrablock. Didn't help (at least not significantly).

Floating point...I don't even have the option to turn it off.

Disabling "no motion search for still picture" didn't help anything.

I haven't tried different quantize matrices because I know nothing about them. I don't have any idea what those numbers mean.

[mikehg]

Actually I've had this problem on occasion and using Svcd instead of vcd seemed to help reduce it. From my experience, it seems like reducing resolution causes alot of these artifacts. So it may be worth a try to encode a short clip in svcd and see how it looks.

Good luck,
Mike H.

[rbruner2]

I can't encode to SVCD now...my time has expired. Anyhow, I wanted to get this to work for VCD if possible since some of my family has VCD compliant DVD players...but not necessarily SVCD compliant.

I wish I could try SVCD now, however. I'm holding off purchasing Tmpgenc Plus for now in hopes that Pinnacle actually improves their MPEG encoder in Studio. (BUt, they probably won't, so in the end, I'll end up buying Tmpgenc Plus.)

[kinneera]

My suspicion is that VCD simply is below your quality standards. And I'm still a bit curious about your TV, since its very atypical for it to look better on a computer screen.

[rbruner2]

VCD quality is actually fine if it weren't for these artifacts surrounding the edges of things. But, I have gotten the same problem on SVCD's I've done in the past.

If you want to see the problem, do this: Take any video with some action in the background. (There doesn't even need to be a lot of action...just some motion.) Then, using an editor, overlay some title text that is stationary.

Now, convert the resulting movie to VCD using tmpgenc. Now, look around the edges of the text, and you'll notice the artifacts I'm talking about...it is like there are blocks in the background where there is supposed to be motion that are not moving. In my particular case, I have an "interview" style scene. THe camera is stationary aimed at the subject's upper body while they talk. This essentially makes the background completely stationary, since the only thing moving on screen is the person, and even then, very little movement. THe result is that there appears to be surrounding the person (around the edges of the face, etc.) portions of the background that sort of move or look distorted as the person moves. If the person is perfectly stationary, of course, there is no problem.

Regarding my TV. It is nothing special. A 27 inch Magnavox that is about 4-5 years old. (I don't know exactly, since it was my parents first, and they passed it on to us when they moved.)

But, I've noticed this problem on other TV's as well.

There are other problems that show up more on the computer screen, such as obvious edges where colors change that, on TV, look more gradual. But these particular artifacts are small enough that viewing them on a monitor in high resolution seems to make them less visible than when magnified to 27 inches.

[kinneera]

VCD quality is actually fine if it weren't for these artifacts surrounding the edges of things.

Well, that's exactly my point - the artifacts bother you too much. The suggestions people have given are about all you can do with standard VCD - and if it still bothers you, then VCD isn't the format for you. Wait for a DVD burner.

But, I have gotten the same problem on SVCD's I've done in the past.

SVCDs are a whole other story. The flexibility of 2x the video bitrate in combination with variable bitrate encoding mean that you should be able to achieve SVHS quality given a decent enough source (I can attest to this being possible). And unless you are encoding animation, you can generally eliminate Gibb's artifacts for all practical intents and purposes. You have to make the decision on whether you're willing to commit to SVCD, though, and then really commit to it because it will take some effort to learn how to get the best results.

[rbruner2]

Why is it that VCD looks good for high motion stuff, but for little motion stuff like this, when you shouldn't need the bitrate, is when quality suffers?

[kinneera]

I would guess this is more pyschological than realistic...the artifacts are almost certainly worse during high motion, but the motion itself distracts your eyes (and also tends to cause the artifacts to move rapidly) such that you find them less bothersome. Typically high motion causes macroblock artifacts, not Gibb's artifacts.

A static scene gives you plenty of time and eye boredom to notice the artifacts. And once you've noticed them, you're also somewhat preconditioned to recognize them, even if you've succeeded in substantially reducing them. It also doesn't help that you're trying to encode text overlays at VCD bitrates - the hard, sharp, and actually very complex shapes of letters will require a fairly substantial bitrate to reproduce without such ringing artifacts.

[rbruner2]

Macroblock artifacts don't bother me as much. What you're saying is correct about the psychological aspect of it...I'm just surprised there isn't any filters or anything to compensate for this. Frankly, I would rather have less quality in the overall image, with fewer artifacts. The artifacts are the very first thing I notice in any VCD and SVCD. Even some captures that my neighbor has done at high bitrates I tend to notice them. (I'm talking 3000+). He actually doens't even create VCD or SVCD standard or non because his Apex DVD player will play just MPEG files copied to a data disc.

That isn't an option for me because my whole intent is to distribute them to other, so I need everything to be as compatible as possible.

On a side note...in order to be an SVCD, does it have to be encoded at MPEG-2? In other words, is there any difference whatsoever between MPEG-1, and using a fixed bitrate MPEG-2?

I could create a 480X480 resolution file using 2400 as my bit rate to see if that helps, and burn it as an SVCD. But I wasn't sure if there is something inherantly different between MPEG-1 and -2 other than the fact that MPEG-2 supports VBR?

[kinneera]

Yes, by definition SVCD is MPEG2. My strong recommendation, if you are sensitive to artifacts, is to also use 2-pass VBR rather than CBR.

As for the differences between MPEG1 and MPEG2, well...it's been discussed quite a bit previously and I encourage you to search for previous threads if you're curious about the topic. The short of it though, is that given equal encoding parameters, there is minimal technical differences in quality. However, various (software) encoders tend to perform much better at either MPEG1 or MPEG2 but not both. MPEG2 also supports a number of broadcast/DVD oriented features, which are not really important to your discussion. Also note that MPEG1 supports VBR, but there is no standard based on it.