Average out brightness.

[MarkYoungIW]

Hi,
I am new to any sort of video restoration, and very very impressed at some of the tools I have used so far.

Most things I have tried to do, I have found guides on the net and in general been very impresssed at the outcome.

I have a short clip that I need to stop the change in brightness (which unfortuantely I think has occurred when the camera has panned to the sky). I am using the clip in a DVD menu, and it would be the perfect clip if I could stablize the brightness.

The clip is 16 seconds long, and currently saved as an uncompressed .AVI (485mb). I did not do the capture from VHS to DVD, so this has been taken directly from the .VOB files we have, and saved as uncompressed .AVI with no processing.

To give an idea of the problem, I have uploaded it to YouTube. If anybody can help, I have also uploaded the full file to Uploading.com.

I would be very grateful for any help. I have mainly used VirtualDub and Sony Vegas so far. I looked at AVISynth, but I couldn't work out what I was meant to do with it, but I have seen mention of AVISynth filters that may help.

Thanks in advance.

Mark

[sanlyn]

Your AVI might not be that useful, as we don't know how the VOB->AVI conversion was done. You say you didn't capture from VHS to DVD, so where did the VOB come from and how was the AVI generated? I'm afraid your narrative is somewhat confusing about which version came first, AVI, VOB, or DVD, and what format was used for the original camera shot? You would have to work with the original, unprocessed camera recording (unless it's tape, of course), or as close to that as possible.

Your camera tried to record a chroma and luma range that is far beyond the capabilities of analog or digital video -- or film, for that matter. Still, you might be able to salvage some of the levels and details. But note that at one point your trolley goes entirely black.

Your link to uploading.com just gives a "file is still uploading" message. You can reduce the size of that AVI to about 1/3 its current size with Lagarith lossless compression in VirtualDub, using "fast recompress" to avoid colorspace conversion and setting color depth for the AVI's original colorspace (which should be either YV12 or YUY2). But we still don't know how you derived that AVI.

[MarkYoungIW]

Sorry, I will and try and explain a bit better. Basically the VHS->DVD conversion wasn't done by me. All I have is the DVD.

I took the relevent VOB from the DVD straight into VirtualDub. Selected and deleted everything except the clip which I then saved as AVI with no compression.

Hope that makes a bit more sense. Not sure why Uploading said that. Seems to be working now, so it was perhaps still processing or something techie!

Unfortuantely I do not know what the original camera source was, or what was used to capture the video and make it into a DVD.

Hope this clears my situation up a little more.

Thanks again

Mark

[poisondeathray]

To edit the vob, you could have used something like mpg2cut2 . The filesize would have been 10x smaller with no quality loss

I can see the problem in the youtube video that you're likely facing - parts of that video have a dark/bright "split" . You don't have even luminance distributed through the frame. So any "automatic" levels adjusting or averaging plugin will only see that as "average grey"

For example if you have 1/2 pure black and 1/2 pure white . The average will be grey . So a typical averaging plugin will do nothing (or not very much)

There are "windowed" histogram equalization plugins, that adjust different parts of the picture , differently . There is one for vdub you can search for that

But I suspect what you're going to have to do to get a satisfactory result is adjust it manually with keyframes - you brighten or darken different parts of the picture and adjust it over time . You should be able to do this with vegas pro with the curves or levels plugin

[sanlyn]

OK, thanks for that info. Makes more sense now.

So, recording VHS->DVD directly was not the best idea (it's one of the worst, in fact), but the DVD is what you have to work with I guess. When you opened the VOB in VirtualDub the YV12 original was converted to RGB32, most likely with a subsequent loss of detail thru crushing of darks and blowing-out of highlights.

The conversion of VOB/YV12 to AVI/RGB32 or AVI/RGB24 has to be more controlled. There's a way to do it properly in Avisynth. But you need to work directly from the original VOB -- before any other processing or conversions. The original VOB is the best (maybe only) way to help clean up the problem.

So to get a sample of the original scene from VOB, I invite you download a copy of the free dgmpgdec utility (see this link in the forum's Tools section: https://www.videohelp.com/tools/DGMPGDec ). The download will be a ZIP, so make a folder for that zip file and call the folder DGMPGDec. When you unzip the download you'll see a file named DGIndex.exe and a DLL named DGDecode.dll. Leave the DLL in the same folder with DGIndex.exe.

Copy your VOB to a folder in your hard drive. Double-click DGIndex.exe to run it. In the DGIndex application window click "File" then click "Open". Locate your VOB file, then click OK. After a few seconds the VOB will open in a video preview window. Note at the bottom of the window are scroll controls. There are two buttons labeled "[" and "]". The [ button is pressed to mark the beginning of a cut of your video, the ] button is used to mark the end frame of the cut. Note that the scrollbar on the bottom of the window is not continuous; it stops on key frames only. Mark the start and end of your desired cut. Then click "File" and click "Save project and demux video". The resulting edit will be an .m2v file (think "MPEG2") placed in the same folder with your original VOB. There will also be a .d2v file, which you won't need for now.

Open a new reply in this thread and use "Upload files/Manage attachments" to upload the .d2v. You won't need the audio. The max file size is 100MB, but this short scene won't be that large.

You will need a copy of Avisynth 2.5. Get it here: https://www.videohelp.com/download/Avisynth_258.exe

This problem can't be fixed in RGB. It can't even be analyzed properly without the original VOB.

[MarkYoungIW]

Hi,
Many thanks for that. Hopefully I have got this right.

Looks like I've got a lot to learn. I only ever remember using avi for lossless video, but obviously things are a lot more complicated than I thought!!

Thanks for the help with this. I've been looking round the forum. Very very interesting, but at the moment an awful lot of it goes straight over me!!

[sanlyn]

Thank you. Doing OK so far.
I'm certain a few dozen members will jump on this one.

[sanlyn]

THE VOB and the AVI are both progressive. The luma+chroma range is a little less severe than in the AVI (although there's one point where a corner of bright sky is completely blown out. Oh, well, that happens).

I'm considering a few ideas, but I note some strange disturbances. Anyone notice something odd about the red noise and flicker in the tram car?

[MarkYoungIW]

THE VOB and the AVI are both progressive.

Like I said, I am a beginner at this, but I would take it as VHS is interlaced it would have been better if the VHS had been captured in interlaced as deinterlacing is likely to have generated additional objects? Although I won't be able to track down the VHS before the DVD is needed (Christmas!) I am guessing it would be a worthwhile thing to do for future restoration.

[sanlyn]

The manner in which it was deinterlaced has, I'm sorry to say, created insurmountable problems. The frame I posted earlier illustrrates one such glitch, but it gets "glitchier". The dark areas mask distortion. It's possible to brighten some things, make a few corrected copies of the clip, divide the scene into 2 or 3 "zones" of varying contrast and then dissolve in Avisynth from one to the other. But the noise is terrific. The video was deinterlaced incorrectly -- frames (or fields) were discarded, so much of the original data is missing.

Another problem is that crushed or compacted detail is noisy. There isn't enough pixel info to differentiate fine contours. Expand the dark area, and the "missing" info just becomes noise (which is what it is).

Expanding the dark smudgy stuff reveals what's really there. Rather ugly. If you load one of the "dark" frames and look at lines and corners more closely, you'll see the distortion and mottling without expanding those areas. I used contrast-reduction masks; these dandy masks can be configured to reduce contrast variably, expanding values in the darkest and middle areas while retaining highlight detail.

Example: Look at frame 95 in your original VOB. The shadows reveal more than you first suspect: look at the notched and wrinkled edges of the dark lower parts of the tram car, and at the black bars upper left. The original interlaced fields contained the data that made those lines look smoother in the VHS. With improper deinterlace, the data that was retained is only "half" of what was recorded.

frame 95 (original vob):

[Attachment 15134 - Click to enlarge]

frame 95 (shadows expanded -- not yet optimized, but used as example)
A look at the time/date characters reveals something else the darkness hides, as well.:


[Attachment 15135 - Click to enlarge]

These disturbances differ from frame to frame, looking worse in motion. You can load the m2v directly in VirtualDub, and then view it one frame at a time with the < and > buttons. Observe the noise frame by frame as the tram rises.

I began by denoising and trying to repair a few artifacts, working first on the dark ending. Masking and some color and some VirtualDub level corrections and color filters made an improvement. Even in a Hollywood film you often see objects darken against a bright sky (it's preferable to having the sky a complete wash-out!). But there's no way to repair the distortion and the split lines on edges.

frame 469 - original:

[Attachment 15136 - Click to enlarge]

frame 469 - correction attempt:

[Attachment 15137 - Click to enlarge]

I'd say this clip would be a poor choice, as is. For informative purposes, I've attached the script I used to denoise a bit and apply the contrast mask.

[themaster1]

something is strange about this VOB it is PAL yet it use the smpte 170m matrice which is made for ntsc, PAL should be BT.470-2 B/G

[sanlyn]

Yes. A host of strange happenings here. IMO starting from scratch is the only way to go. I don't know where MarkYoungIW lives (profile doesn't say) but Jasper Tramway is in Alberta, Canada. Doesn't Canada use NTSC? I forget.

[themaster1]

if i had to guess i'd say he's used a multi standard dvd recorder and wait for it.... bought in north america -> hence the default smpte170m matrix used (i'm guessing engineers didn't pay much attention to that ?)

[sanlyn]

Sad to say I'm not familiar with the hardware, but O.P. said the DVD transfer was done for him. Even at that, the bitrate is too low for noisy VHS transfers. Supposing the frame structure was correct and intact, that's still too low.

[MarkYoungIW]

I'm Isle of Wight, UK. And I am fairly sure it would have been a PAL Camcorder. I am thinking you are probably right that I should get the VHS recordings and start from scratch. Its strange actually, I noticed very few of the glitches in the video when I first looked, but now I'm looking for them I keep spotting them!! I think you probably right. For now I will look for a clip in better condition to get the 1st release of the DVD together as I am currently pushed for time, and then recapture everything again when I'm not so pushed for time.

I have an SVHS player, but currently no hardware for the capture. I am guessing capture devices will make a lot of difference to quality as well?

[smrpix]

For now I will look for a clip in better condition

You would be surprised how many people never think of this solution. My hats off to you.

[manono]

I don't believe it's a direct capture to a DVD recorder. It would have been encoded as interlaced. It was reencoded later (where the crap deinterlacing took place), perhaps in TMPGEnc I'm thinking, as I believe SMPTE 170M is the colorimetry it uses. I could be mistaken, though, as I don't own TMPGEnc and can't test.

I do have PAL DVDs in SMPTE 170M and NTSC DVDs in BT.470-2 B,G, so I don't believe there are any PAL/NTSC issues here.

[sanlyn]

I didn't see anything in my TMPGEnc having to do with SMPTE 170M one way or the other, but we don't know what edition of TMPGenc was used. Mine is ye olde TMPGenc Plus 2.5, I never used later ones but went HCenc when I want a newer encoder. That deinterlace is a shame, it looks as if the O.P. does OK with setting up the camera shots -- that pesky auto exposure aside (I can't stand autogain. It never does what you want it to).

Never fear, MarkY. There's always tomorrow. I don't have time to finish my Christmas project either, darn it. Not even close.

[sanlyn]

I'm Isle of Wight, UK.

Lovely area, I hear. I didn't have time to get there on my two UK trips. The wife wasted too much time with those damn tourist buses. Maybe next time.

I have an SVHS player, but currently no hardware for the capture. I am guessing capture devices will make a lot of difference to quality as well?

Certainly, keep the SVHS player. A dedicated capture device would make a vast difference, but research that area first. There is huge variation in quality. Mine are 8 to 10 years old, so I can't advise on that, and they are NTSC. But there are several good ones available.

[manono]

I didn't see anything in my TMPGEnc having to do with SMPTE 170M one way or the other

If I'm correct (and there's no guarantee I am), you don't choose it. It just uses it and you can check afterwards by opening a VOB in DGIndex and running the Preview.

[sanlyn]

Will check it out, thanks for the tip. I learn something every 15 minutes around here.

[manono]

OK, good. One more thing. You can't always get the colorimetry from DGIndex. If there's an asterisk, it means the colorimetry wasn't stated in the file and DGIndex just shows 'BT.470-2 B,G*'. That's the case a lot of the time. The chances are very good it really is BT.470-2 B,G or equivalent, but there's no guarantee of it.

[themaster1]

Make an mpeg2 dvd ntsc export with pro coder (no matter the version) and see what matrix it uses in Dgindex or DgindexNV , it's gonna be a fun experiment
Now, compare the colors with your original video and you will understand why the colors are wrong (well probably not the only factor)

When i tell you to use mainconcept ref2 i hope it's not fallin into deaf ears

[sanlyn]

Well, no, it might not fall on deaf ears but it will surely fall on an empty wallet. I'm still paying for my AfterEffects.

[manono]

Make an mpeg2 dvd ntsc export with pro coder (no matter the version) and see what matrix it uses in Dgindex or DgindexNV...

Why not just tell us? I don't have either of the two encoders you mentioned. I'm happy with CCE. My understanding is that unless stated explicitly, DVD video gets decoded as BT.470-2 B,G. If it's something else (and says so) it should still be decoded correctly. It's only if the colorimetry isn't stated and it's something else (or if it's stated incorrectly) that the colors come out slightly wrong.

[themaster1]

ITU-R BT.470-2 System B, G = PAL ( Rec601)
ITU-R BT.470-2 System M = NTSC / USA ( Rec601)
Source: http://scanline.ca/ycbcr/

Read the 4.1 Differences in RGB chromaticities

[manono]

Thank you. Are those choices you make in MainConcept, or are they what DGIndex or DGIndexNV show. I assume those are what you choose in MainConcept. What does DGindex show? Not exactly the same as what you wrote, I don't think.

[themaster1]

There are two matrix for ntsc Sd sources it seems (if you have read the page) ITU-R BT.470-2 System M and smpte 170m, mainconcept use the smpte170m.With Hcenc you can use either the BT 420-2 M or smpte 170m matrix

Any pro encoder that use the 470-2 BG matrix for ntsc dvd was made by sloppy devs i say

[poisondeathray]

But they are just flags - the YUV data isn't changed no matter what flag you use. Many decoders ignore them anyway

Procoder is a special case - if you don't use a specific workflow and use YUY2 input you incur RGB conversion and levels shifts

[2Bdecided]

Sanlyn, you expect perfection! Really, it's not that bad. At its very best, it's still going to be home movie footage shot in auto mode (=high contrast, blown out video) on an old camcorder. Hopefully a transfer from the original VHS, using a time base corrector (see how wobbly the date is) will avoid some of the noise.

People watching it are going to be either excited or comatose by the content, not the quality. It would be nice to get rid of those noise bursts though.

Cheers,
David.
P.S. looks nice... http://www.jaspertramway.com/