How do I rectify motion blur on mkv's

[Acehit]

What do you mean I don't get it? The final result from handbrake leaves the original aspect ratio in tact. Please explain.

[jagabo]

By converting you AVI first you've probably lost the pixel aspect ratio information. Handbrake is probably assuming square pixel, making the full frame 3:2 DAR, not 4:3 DAR. So your final video is about 9 percent too wide. This is approximately how it should look, assuming your players scales correctly for the PAR flag...

[Acehit]

Not at all. I tested both the original and the final output side by side and aspect ratio is identical. I even tested both on my wd live media player connected to my tv and the frame is also identical there too. On handbrake I select 'strict' and auto-crop.

[jagabo]

Post a sample of your final output.

[Acehit]

Hey wait a minute no my mistake, I apologise you are right! I just assumed that the result would match the first output from handbrake. Blimey how did that happen!

How do I fix this? Is there is a additional script I can put into the script to rectify this?


The problem with virtual dub is that it will only allow me to output at selected aspect ratio's. You can't select a custom aspect ratio.

[sanlyn]

The original movie was 1.875:1, not 16x9. The mkv sample originally posted showed a 4:3 frame with a 1.8:1 image letterboxed into the central "active" 704x480 4:3 frame area, letterboxed and with border pixels on each side. That image is slightly too narrow: it is neither 16x9 nor 1.875:1. You cannot entirely fill a 16x9 screen with a wider 1.875:1 image -- the wider image, if it is of the correct proportions, will be letterboxed in a 16x9 window or TV.

The video I posted will play as a 1.875:1 image inside a 16:9 frame. There are 12 pixels of letterboxing at top and bottom to preserve the image aspect ratio. The script I used for denoising and resizing is in post #51. The movie I made from that script is attached in post #18. Play that clip full screen and you will see that on a 16x9 screen it is slightly letterboxed. If you crop off the pixels, the image will be stretched vertically, which is where you started.

You are encoding to a 720x480 standard definition encoded frame size. 720x480 encodes don't use square pixels; for 16:9 encoding, they use rectangular pixels that are defined to a player as being higher than they are wide. Thus, the player will stretch that image horizontally but not vertically, to make the 1.5:1 720x480 frame into 1.777778:1. With the top and bottom pixels cropped off, the image inside that 1:777778:1 played image will not be a 1.875:1 image.

[Acehit]

Thanks sanlyn, you are right my understanding of aspect ratio's isn't that comprehensive and as I read your last post I am using up my full brain power and still I am trying to understand. I think I'm better off just mirror matching it in virtual dub using the resize filter and just match it to the original. I know you said the original is wrong also, but it is what the original dvd is, so I i'll just settle for that.

[sanlyn]

That's up to you. If I know the original is incorrect (and quite often they are), I don't see that correcting is such a terrible ordeal -- especially when you already have a script that does it for you. I'd suggest that Avisynth has superior resizing methods than VirtualDub. The purpose of restoration is to restore to original or better quality, not to duplicate errors that can be corrected.

But, then, the purpose of this video hobby is to get what you want, not what someone else prefers.

[Acehit]

I do get what you mean, but the problem I have is 'how do I know what the "actual" aspect ratio should be? I could be here all day trying out different ratios trying to figure out which is correct.

[jagabo]

I don't know where Sanyln got 1.875:1, but if you take the original 720x480 frame and crop away the black borders you're left with about 698x343. Since this is obviously from an analog tape recording I would assume the standard rec.601 pixel aspect ratio of 10:11. If you do the math you get (698/343) * (10/11) = ~1.85, one of the standard widescreen movie aspect ratios.

If you don't think that is correct you should look for something in the movie of know aspect ratio and measure it on-screen. A sphere, the wheel of car viewed directly from the side, etc. But beware, lenses often have distortions near the corners of the frame so look for an object near the middle of the picture. And large enough for you to measure accurately.

[Acehit]

That's the problem, I could be here forever trying to do that. Normally I just go by the shape of the heads and bodies, if they look too wide or too narrow than I just go by eye, but to get the precise exact ratio to 100% I don't know if its possible.

[jagabo]

In my opinion, one or two percent error in aspect ratio is ok.

[sanlyn]

I got the aspect ratio from imdb and an Indian movie site that talked about the director's work. And look at the info on the "Certificate" when the film opens.

[Acehit]

Guys I have just realised something maybe you can shed some light on.

1. The framerate of the original video is 29.97 fps. Now when I put it into virtualduv via the scipt, under video > framerate setting it says the original framerate is 25 fps. I don't get it, does this mean it's losing framerate information as well as aspect ratio data?


Now here's the funny thing, if the original is 29.97 fps and virtualdub outputs the video at 25 fps then in theory the audio should be out of sync right? Well I did a test in nandub, since virtualdub doesn't accept ac3 audio files, and the audio is in sync all the way through! How is this possible?


2. Another thing is to encode the entire video with yadif is going to make my computer 20 hours to complete! Is there anyway I can speed this up? Or is it a simple case I have to upgrade my ram and processor? Just out of interest, I initiated a direct stream copy, and this would've only taken 6 hours but would've given me a 120 gb file size!

[poisondeathray]

1) You're not losing anything except hopefully the blends. Because it's been field blended to 29.97 from 25. There are extra frames added from the field blending to make up 29.97 from 25 in the first place (that's why there are inserted blurry frames). You're just attempting to remove those that were added. The running time is the same, the audio is the same when you reverse the process with srestore . The reason they did this is to make it compatible with NTSC standards

When you use avisynth / vdub, it's assumed square pixels 1:1 . When you encode, you set the AR during encoding in the bitstream, or in the container afterwards, or another method is to resize to square pixels equivalent dimensions (e.g. you would do that for things like web streaming)

2) If it takes 20 hours vs. 6 hours for the same script for uncompressed (direct stream copy) vs compressed, the difference is the compression encoding settings you 're using. You can use faster, less compressed settings or slower , more compressed settings. Describe what settings you are using

Upgrading hardware will help for sure either way.

[Acehit]

I guess I could output to xvid which will take shorter time and then convert back to mkv in handbrake. Hopefully there should be no visible quality difference by doing this by the time I get to the finished file.

[poisondeathray]

I guess I could output to xvid which will take shorter time and then convert back to mkv in handbrake. Hopefully there should be no visible quality difference by doing this by the time I get to the finished file.

not a good idea , lower quality, slower

what are you exporting right now from vdub ? using the vdub external encoder ? x264.exe ? x264vfw ?

[Acehit]

well the encoder I am using is x264 - H264/AVC Encoder. On save as, I select all files, and add extension .mkv.

I am also using the resize filter.


This is what takes 20 hours to complete. I'll check for another, faster encoder.

What do you suggest?

[Acehit]

Hey I just experimented with another codec called microsoft mpeg 4 video codec v3, and it takes 9.5 hours with a video rendering rate of 7.5 fps compared to the previous 3-4 fps with the H264 codec.

Is this the best way to go? At least I am outputting to mkv rather than avi. I also experimented with xvid HD codec, that also allowed me to output as mkv, but this took longer and predicted a huge file size.

[jagabo]

Why don't you just use faster settings in the x264vfw encoder? It will still look better than MS MP4v3 at similar bitrates.

[Acehit]

Hey you're right. I've discovered that there is a setting called single - pass bitrate, I will use that instead of single - pass quantizer. I didn't notice that before I just experimented with it and it is much better. The predicted overall time is 17 hours at 5000 maximum bitrate setting, which I can just about live with, and the projected file size is exactly what I want it to be. Perfect I have now everything in place ready to fix all my ripped movies. A massive thank you to all of you for the priceless advice and guidance!

[poisondeathray]

You can use faster settings if you want. I think the defaults are "medium" . Hanbrake defaults to something faster, unless you select "high"

[jagabo]

There are a few different builds of x264vfw. One has a pulldown called Preset in the top left corner. It ranges from Ultrafast to Placebo with eight steps in between. I wouldn't go faster than Veryfast for general encoding but there is a 100 fold difference in speed between the Ultrafast and Placebo settings. And several fold difference between Veryfast and Medium.

Note that the difference in encoding speed may differ if you performing a lot of filtering -- because the filtering won't go any faster.

Handbrake's Normal preset is equivalent to x264vfw's Veryfast preset.

[Acehit]

Guys, I have just started encoding another movie, and I have encountered something I never experienced with bol radha bol. This time the aspect ratio of the movie is 2.35:1.

Take a look at the screenshot, when I import and ouput the movie in DGIndex, it narrows it for some reason. So by the time I put into virtual dub I have a picture that is shorter by width. On Bol radha bol, it only changed by height, which was easy to fix. But is there anyway of outputing from DGIndex so that it maintains the proportion of the movie?

[Acehit]

sorry its here.

[manono]

Take a look at the screenshot, when I import and ouput the movie in DGIndex, it narrows it for some reason.

No it doesn't. DGIndex is showing you the way it's stored on the DVD, 720x480, just as Bol Radha Bol. In this case the DAR is 16:9 whereas Bol Radha Bol's was 4:3.

But is there anyway of outputing from DGIndex so that it maintains the proportion of the movie?

What's the intended final output? If another DVD you keep it at 720x480 and encode for 16:9. If you want square pixels for something then crop away the black and resize to something like 640x272 maybe.

[Acehit]

Guys, check this screenshot. This new video is a much better picture quality then the last one as its a more recent movie, but during scenes of intense motion, the blur that you get from de-interlacing or whatever it is has reduced to this. A very faint edge halo. Now I could use qtmc but since I do not have the horse power to run that, are there any additional filters I could use to eliminate this? If not it's no big deal, you don't notice it anyway, but if there is an easy way to get it 100% clean then I'd like to know.

[manono]

Why do we have to tell you again that pictures are next-to-useless? Post a video sample. 10 seconds with steady movement will be plenty.

[Acehit]

Here it is, the reason why I didn't upload this initially is because with a screenshot I can exactly show you what I mean.

[jagabo]

Is that the source or your conversion? You need to provide a sample of the source if you want help figuring out what's wrong.