Test captures of VHS tapes with JVC HR-S9500

[knumag]

If you download the test.avi from the link https://drive.google.com/folderview?...DA&usp=sharing
you can see my result. I feel there are some strange shadows when there is movement on the picture.
Might this have something to do with the "BEST" setting on the vcr?

Thanks in advance

[lordsmurf]

There's nothing at that link.

Edit: All attachment from posts on page 1 are available here:
NOTE: Right-click and 'save as' the links -- don't click the link and load them in the browser!

http://cdn4.digitalfaq.com/knumag/test.avi
http://cdn4.digitalfaq.com/knumag/eks1.avi
http://cdn4.digitalfaq.com/knumag/eks2.avi
http://cdn4.digitalfaq.com/knumag/eks5.avi
http://cdn4.digitalfaq.com/knumag/filt.avi
http://cdn4.digitalfaq.com/knumag/filt2.avi
http://cdn4.digitalfaq.com/knumag/unfilt.avi
http://cdn4.digitalfaq.com/knumag/unfilt2.avi

[knumag]

Didn't expect you to check so fast finished uploading now

EDIT:
You should download, playing it in the browser makes the quality worse I think.

[lordsmurf]

I was online.

Google screwed something up. All I can see are a JPG and Flashy files. It's not original.

You're premium. Go get FTP access: http://www.digitalfaq.com/forum/news...mbers-how.html. Uploading to FTP removes the headaches of large samples. You can also attach files to the forum. Either 32mb max, or multi-RAR 32mb chunks.

[knumag]

You have to right click and download the file, its just google stupid previewer.

[lordsmurf]

Again, all I can get are a JPEG and a converted Flash file version. There's no AVI.

[knumag]

Strange it works for me, even if not logged on to my google account.

https://drive.google.com/uc?id=0B_UC...xport=download

[kpmedia]

Quote:

Originally Posted by knumag

https://drive.google.com/uc?id=0B_UC...xport=download

This works.

I'm downloading it to the local net. We've got it now.

[knumag]

So? or still downloading? what do you think about the capture?

if you want me to make the file smaller or something else, please tell, I'm eager to hear your advice I'm situated in Norway and doing this as a part time job, digitizing cassettes for people, mostly its video8, Hi8 and D8 though, people probably think they can digitize vhs themselves with their old VCR. Trying to get a hold of a Datavideo tbc1000, but hard to get my hands on, hope the VCR does the trick, also waiting for one MR-DH30 that I purchased, for the LSI-chipset. This recording though, is simply the VCR to Dazzle DVC-100 trough s-video with virtualdub captured as YUV, then re-encoded to matrox-DV and resized to cut out the top bottom side stuff, no other filters used.

With regards

[sanlyn]

Hmm. Well, if no one will comment I might as well give it a try. Other viewers might have different impressions, but the following are mine. Hope you don't take this the wrong way, but there is no way to avoid mentioning several major problems.....

Your sample is a bad capture that can't be repaired. It's soft and badly blurred, even for tape. Fine detail has been removed to the point where the tape looks denuded, and the player's edge enhancement and other noise reduction along the capture chain give it a plastic look with visible posterization and clay-face effects. There's bad motion smear and ghost trails (which you called "shadows"). Tape normally plays as interlaced, top field first; this capture has been unnecessarily and improperly deinterlaced, then encoded as bottom field first. The source appears to be film based and looks as if it was telecined (telecined video should not be deinterlaced). There was no line tbc in use, so you see ragged and noisy edges (a tbc-1000 or other full-frame tbc will not correct this during playback). The images are covered with a fine mesh of thin vertical stripes, and improper deinterlacing has resulted in visible aliasing and detail loss. The motion smear and ghost trails could have been caused by the player, the capture device, or strong temporal denoising, or all of these. Cropping noise from borders and then resizing the video is not the proper way to handle problems like head-switching noise, especially if other uncorrected problems exist in the image; depending on how the resizing is done it can distort the original frame's aspect ratio.

Unless there is some critical processing that requires progressive video, avoid deinterlacing interlaced, telecined, or similar sources. PAL DVD and some HD delivery formats are usually interlaced except for PC-only or web-only playback -- and even then, there are proper and improper methods for doing it. If the source tape is digital, it should be copied to a computer (not recorded, as you suggest) via Firewire to DV. Avoid capturing analog sources to DV.

If the original tape was a slow-speed recording, the JVC is not a good candidate for playing such tapes. I would say that you have to learn more about recognizing video problems and their causes and solutions. You also need a better capture device than anything from Dazzle, especially if you want to charge clients for work on their videos. Your statement that the original tape was a "well recorded cassette" isn't evident in your sample, so there is little that can be determined about the original.

That's my personal take. Others might have a different ideas.

[knumag]

I just added two different parts from the same cassette, dont really know what to do with them, I have a minidv camera, better to use that one as a digital pass-trough than the dazzle. Seems like the cassette and the recording is the main problem here.

https://drive.google.com/uc?id=0B_UC...xport=download

https://drive.google.com/uc?id=0B_UC...xport=download

What I have done is capture to huffyuv without any filters. I dont know if "Video -> Noise Reduction -> On" counts as filter in Virtualdub. I then resize it with Virtualdub and crop the bottom, to remove the blurred line, and sides ( I capture to 480x576 and resize to 720x576), and check the box that says interlaced. I use the codec Matrox DV/DVCAM for the compression. The Virtualdub I'm using is the one I got on this forum.

Am I doing anything wrong?

[sanlyn]

Sorry for the delay in replying.

Quote:

Originally Posted by knumag

What I have done is capture to huffyuv without any filters. I dont know if "Video -> Noise Reduction -> On" counts as filter in Virtualdub. I then resize it with Virtualdub and crop the bottom, to remove the blurred line, and sides ( I capture to 480x576 and resize to 720x576), and check the box that says interlaced. I use the codec Matrox DV/DVCAM for the compression. The Virtualdub I'm using is the one I got on this forum.

Am I doing anything wrong?

You are, yes. After seeing your notes, no wonder the caps look strange. From your over filtered posts it's impossible to know how the original tape plays or what it actually looks like. The latest samples are a little better, but for a few seconds I thought I was looking at computer-generated animation. That's not an exaggeration.

You are capturing a 480x576 image (1.2:1 aspect ratio) . Most would capture PAL at 720x576 -- depending on what your capture device allows. From most VCR's you'll have about 12 to 16 pixels of side borders and 8 to 12 pixels of bottom head-switching noise. Rather than doing a lot of heavy resizing (which you shouldn't be doing in VirtualDub anyway, as there are better methods), most people crop off the black border and noise and replace those with black borders to retain the original image proportions. Advanced users would employ basic Avisynth for this using Avisynth's Crop() and AddBorders() functions to restore and center the image within the output frame. However, you're cropping in RGB with VirtualDub, which is a color conversion you might or might not want (or need), and employing a destructive form of deinterlace. The resizer used is apparently creating unpleasant over-shaparpening artifacts.

If the VirtualDub filter dialog is telling you that denoising is "on", it is most certainly "on". Yes, it does count as a filter. The same goes for your capture device. That denoiser is doing some damage and creating new problems as well as slowing down your capture and performing unwanted color conversions from the original YCbCr to RGB and back to YUY2. VDub's resizer is using field decimation to deinterlace (which is costly and results in poor motion handling with PAL playback). There are far better resizers and deinterlacers if you need them. My question is, why do you need them? Then you are encoding to DV-AVI, not a very friendly color matrix for VHS source. Again, this means that we have no way of knowing what the original, unprocessed capture looks like. The two newer posts have the same clay-face and posterization problems, plus bad edge shimmer, ghost trails, and severe aliasing (the latter caused by improper deinterlacing). Play your eks1.avi video and notice the aliasing and shimmer on edges during motion. Note also how facial contours shimmer and change shape with motion. Don't use VirtualDub's capyture filters. Instead, use the "Levels..." filter dialog, which connects with your graphics card and won't involve colorspace changes. Simple, faster, and less destructive levels adjustments are Brightness, Contrast, and perhaps saturation. Don't sharpen VHS (you'll sharpen noise). Don't bother with color controls because VHS changes color balance from minute to minute and will drive you crazy. Denoising comes later. Avisynth and VirtualDub have far better denoisers.

Let's say you've captured VHS to 720x576 for PAL. If you had 12 pixels of a side border and 8 pixels of bottom-border noise, you would crop off 12 side pixels and 8 bottom pixels. Then, rather than resize the original image, you can restore the original frame and center the image by adding 6 black pixels to each side, plus 4 pixels each at top and bottom. Those black pixels are hidden by normal overscan anyway. Meanwhile, the original image content would still be interlaced and otherwise unaltered.

It appears that you're using a VCR that is a poor choice for home made VHS. The 9500 is a basic budget VCR, not a high-end JVC. Advanced users have more than one VCR, to match the input source. For this type of VHS source. many tapes give significantly better detail with other players, or with higher-end JVC's. The only JVC's I've encountered that perform decently with home-made VHS, especially slow-speed tapes, are the 96xx and 98xx series that are far more advanced and configurable than the 9500. Advanced VCR's with built-in TBC aren't all that perfect to begin with, and finding one nowadays that hasn't been used to death will be difficult. I've gone through 4 high-end JVC's in the past and can testify that home-made VHS and those JVC's were occasionally a poor match, with one tape looking horrible on one JVC and the same tape looking better on another JVC. For slow-speed tapes, Panasonic, Mitsubishi and others are more suitable.

If you want to make a job of these projects you need better equipment. For example, I'm using ATI capture cards with refurbished non-TBC Panasonic VHS and SVHS from the 1996-1998 era, and an old used Panasonic ES15 DVD-R for tbc pass-through into the PC. I also have a rebuilt Panasonic AG-1980 (NTSC only) that is often more trouble than it's worth and sometimes gives poor performance with some tapes, but it's usually the first player I turn to. For DV tapes use a Firewire device with WinDVD for copying directly to DV-AVI.

Hopefully all of this doesn't sound like Greek or Sanskrit. If it does, there is plenty of information available in this forum and others. For now, you're capturing in an inadvisable manner that causes flawed results as well as unnecessary effort that seems to be making the results look worse.

[knumag]

http://www.digitalfaq.com/forum/vide...erly-crop.html - so this is wrong, or am i completely misunderstanding you.

Thanks for your input, turning off the noise reduction will help a lot then. Where am i deinterlacing? Heres another clip from a VHS-C I captured.

https://docs.google.com/uc?id=0B_UCV...xport=download

I really appreciate your help, I am eager to learn

I will capture with my sony dcr-PC110E as digital passthrough from now on and scrap the dazzle.

Video8, Hi8, D8 and minidv i all capture with firewire of course, but i still get the overscan buzz, or how you might call it. I thought the guide i posted before was the way to handle it. But theres a better way? Or did you just misunderstand how I described doing it before? "I then resize it with Virtualdub and crop the bottom, to remove the blurred line, and sides" With this i meant the way it is described in the guide.

Sorry, English is my third language, so sorry for mistakes.

Thanks again for your help, and time.

With Regards.

[sanlyn]

No problem with your English. Anyone who can manage more than one language is doing better than many native English speakers who misuse their own language continually.

If the methods in your links are the methods you prefer, I couldn't advise in detail: I understand the instructions, but I don't use those methods and I disagree with them. I suppose they work for some people. Your eks5.avi is better in a few ways but is badly damaged, as I think you can see when you view it. I can demonstrate the "usual" method that I use and see everywhere. In any case, it's difficult to advise because your sample is not usable. You should submit at least 8 to 10 seconds of your original huffyuv capture, unprocessed, uncropped if possible, not resized, with some motion similar to that in your other samples. Several seconds of unprocessed huffyuv AVI would be a file size of 45 to 50MB.

If your capture is already in that odd 1.2:1 size, or if you have cropped it during capture, you can submit that -- but do not resize it or submit it to any other processing. It would be best to have an original 720x576, 704x576, or even a 640x480 4:3 original, but what you have will do if it hasn't been through any filtering or re-encoding. Below, I describe how you can edit a short, unprocessed huffyuv sample in VirtualDub. You can also do this with DV-AVI captures ....

Open your captured video directly in VirtualDub. Use the edit keys to cut the end and start points of your sample. But don't save the sample immediately. Do not load any filters into VDub's filter list. To save the sample file, go to VDub's top "Video" menu. In the drop down menu that appears, you will see 4 processing modes grouped together: "Direct stream copy", "Fast recompress", "Normal recompress", and "Full processing mode". Select only the "direct stream copy" mode. Do not select any other processing mode. Make certain that "full processing mode" is not selected. You can now save the sample file. The sample will be saved using the same compression and color space that was used to create it. About 5 or 6 seconds of a DV or losslessly compressed huffyuv video would be about 45MB. Or you could submit a few smaller samples that can be strung together.

[knumag]

https://docs.google.com/uc?id=0B_UCV...xport=download
https://docs.google.com/uc?id=0B_UCV...xport=download

This is from mini-dv copying trough firewire, and then the same process as with vhs, to remove bottom.

another example
https://docs.google.com/uc?id=0B_UCV...xport=download
https://docs.google.com/uc?id=0B_UCV...xport=download

Without sound, since they were talking..

Thanks

[sanlyn]

Thank you for preparing the new samples. I think you can see that the latest results are superior to the earlier captures. I'm not in love with tape-to-DV captures, but they are preferable to a VCR and capture device that give poor results.

The two samples use different crop/masking techniques. #1 is a crop/resize that works fairly well (the differences between the original and the resized version are too small to notice, for most viewers). #2 is simple masking. Both have a slight disadvantage, in that some portion of both versions are hidden in the overscan area or, in the case of the livestock shot, top pixels are hidden by the mask. I note that among my acquaintances who are videophiles and classic movie addicts, TV overscan is a feature that they almost always disable on their HDTV. For these viewers, some black borders are preferable to obscuring or altering the basic image.

I can create a short demo to show how these matters can be handled in a different way. But it will take a short while, because I am leaving to travel for a work project for 3 days. For that demo I'll make use of your unaltered captures. Thanks for your patience.

-- merged --

To make a longer story short, this is how the borders were managed in Avisynth: the bottom noise and the top 2-pixel broken border were cropped with the Crop() function. The two side borders were cropped and replaced with black border. The statement from the Avisynth script for both sample videos:

Crop(10,2,-10,-8).AddBorders(10,4,10,6)

This method left the original central image content as-is, except for 2 pixels at the top (which was a broken border anyway). The masking method mentioned earlier covers 8 pixels of the original image at the top, which the viewer would never see. The full-frame resizing method mentioned earlier leaves 10% to 15% of the image hidden by overscan. Taking the usual TV overscan into effect, the Crop + AddBorders procedure leaves most of the original image intact, and the viewer will see more of the original image on TV despite overscan. This avoids deinterlacing, resizing, and RGB conversion.

On the other hand, both video samples have annoying interlace problems, notably shimmer on many edges -- it is most visible in the beach video. Aside from that, the luma levels are out of the RGB 16-240 range almost all the time; highlights are burned away, especially in the beach video. In the livestock images, highlights on the mother are badly clipped to the point where highlights are changing color. In the beach scene, sand highlights are clipped as "hot" colors with no detail and most of the clouds are washed out. Because many of the highlights in both videos were clipped during capture, very little of it can be recovered later.

The edge shimmer on motion is common with DV, and is often the fault of the way the original camera does its interlacing. I repaired a little of it using the QTGMC deinterlacer as a cleaner, then re-interlaced in Avisynth. The shimmer is most visisble in VLC Player and PowerDVD (which are both horrible IMO), but is smoother in Media Player Classic, MPC-BE and Windows Media Player. Levels and color balance are often a personal preference; I used Avisynth to bring the luma levels into proper range (but could not repair most of the original clipping), and some additional denoising was done with VirtualDub's temporal smoother and Camcorder Color Denoise filter on the #2 sample. Controlling levels was difficult because they kept changing through the camera's auto-gain feature.

Both videos have a visible orange-red stain along the right border. I used an Avisynth de-rainbow filter and chroma smoother filter to clean most of it in the beach video, but the stain in the other video was more saturated and very persistent. Some of it was cleaned, but it's still visible in video #2.

Usually many users would not take this much trouble with a video. That's a matter of personal preference. Color Balance, other denoising, sharpening, etc., are decisions for individual users.

Edit: Just received a PM informing me that I forgot to post the Avisynth scripts. Sorry, my bad.

[sanlyn]

Another email, this one asking about the purpose of adjusting levels in the scripts. I'd think the question belongs in the forum rather than email. Be that as it may, the effect of adjusting levels can be illustrated with the following examples. The top image is the original frame 361 from unfilt.avi #1. The bottom image is the result of the ColorYUV and SmoothLevels statements in the script for that clip. Ultimately the final color balance/levels can be adjusted to suit one's preferences, but the functions retrieved some detail from shadows and highlights and gives the image more "dimension", if you will. Some objects look a bit unsharp because of camera motion and interlacing.

frame 361 - original.png

You must be logged in to view this content; either login or register for the forum. The attached screen shots, before/after images, photos and graphics are created/posted for the benefit of site members. And you are invited to join our digital media community.

In the top image, brights exceed RGB 255 -- even the bottom border head switching noise is super-white. This clipping limits dynamic range because clipping destroys bright detail and colors, many of which which can't be recovered after capture. The bottom image has more detail in the sky, water, and sand. It's a very high-contrast shooting situation, so to keep detail across the spectrum you can't have all the shadows and all the highlights at the same time. I thought sky and midtones should have priority, and gamma (midrange) could use more tweaking -- I think gamma here is a little low, and there's still some magenta chroma noise. Others may feel differently. On TV this scene would look slightly brighter than on a PC.

[lordsmurf]

Quote:

Originally Posted by sanlyn

I'd think the question belongs in the forum rather than email.

Always, always, always.

And all of the files have been uploaded to CDN4, since we hate to lose context of the thread if/when the Google Docs disappears.