Why does my captured footage look soft?
 

Hi all,

I am trying to validate my capture setup before I start actually capturing tapes.

My current setup is: JVC SR-VS30 -> TBC-1000 -> ATI AIW Radeon 7200 AGP -> VirtualDub lossless

The TBC-1000 has been completely recapped, and I am bypassing the distribution amp by plugging directly into the S-Video plugs on the TBC-100 card inside.

To test my setup I found a test image online (Original_test_pattern.png). I used ffmpeg to make a short video clip from this image, scaled it down to 720x480, and encoded it to DVvideo. This reduced the quality of the source image a bit as shown (Test_pattern_DV.png). The reason for doing this was to send the video signal digitally to the JVC VCR over the DV connector. The JVC then outputs this signal from the S-Video ports, giving me the highest quality test signal possible at the S-Video connectors.

I then did a capture of the test signal (captured_pattern.png). As you can see the picture is noticeably softer, and the detail in the small lines in the center of the image on the right side were essentially lost. Also, there is a noticeable chroma shift which I am able to correct for in VirtualDub.

To make sure it was not the TBC-1000 causing the softening, I did a capture without the TBC in the chain (captured_pattern_no_tbc.png). The two images look basically identical to me, which is good since it means the TBC-1000 is not hurting the image.

Now, I would have assumed that this softening and loss of detail is an unavoidable affect of the analog connection between the VCR and the capture card. However, I connected the S-Video output of the VCR to a digital monitor with an S-Video input and it is much sharper, and detail of the small lines is visible all the way up until the section on the farthest right (Output_on_display.jpg). So this shows that the VCR is outputting a high level of detail, and that the S-Video interface is capable of transmitting the detail.

This leaves the capture device/software as the only culprit. There are a few SMD electrolytic capacitors on the AIW card which could possibly be causing problems? Or is there some settings in VirtualDub that I am missing that could cause this? I am capturing losslessly with YUV422 colorspace.

Please let me know your thoughts and suggestions.

EDIT: I forgot that the forum automatically compresses images to a very low quality. I have made the images into short video clips. Hopefully the quality is good enough to convey the info.

The file "captured_pattern_no_tbc.avi" is 25fps like PAL but 720x480 like NTSC. Are you trying to capture PAL or NTSC?

The attached files are just png image files that I converted to short avi clips since the forum compresses attached images to a very low quality. The original capture is 29.97 fps NTSC 720x480. The attached files are meant to be viewed as still images. Hope that makes sense.

The .avi is also lossy compressed (H.264), so it's difficult to determine anything.

Since you used a lossless codec, you can use Virtualdub to select 1 frame and "crop to selection". This creates a 1-frame lossless clip.

Thanks for the tip, I am new to using VirtualDub for anything other than capturing.

I followed your method for the two captures and original DV encoded source file.

For the original test pattern image and the picture of the output on my display, I created a lossless video sequence of the images using ffmpeg and then extracted a 1 frame clip using VirtualDub. These clips should be full quality now.

The files you provided confirm that the "Output_on_display.avi" file definitely contains more resolution than the capture files.

The "captured_pattern.avi" shows some checkerboarding, so there's some crosstalk somewhere in the system, maybe the S-Video cables are low quality or maybe there's some noise coming into or out of the power supply on the TBC-1000 or VCR. In good S-Video cables, the two wires are individually shielded from each other, but good cables like that are becoming more and more difficult to find.

The "captured_pattern_no_tbc.avi" may also have some checkerboarding but it's much less.

Also you should use HuffYUV for capture, not RGB.

Okay, I'm glad you agree that there is an issue somewhere and it's not just my imagination.

I am using high quality S-video cables. They have two separate shielded coax cables inside, and the whole cable is enclosed in another layer of shielding. I can post a picture later if you'd like.

If the output looks good on the display, wouldn't this rule out my cables as the issue anyways? However, I was using "Output 1" on the JVC to connect to my capture card, and "Output 2" to connect the display. Perhaps "Output 1" on the JVC has an issue.

I did the captures with YUV422 lossless. The extraction of the 1-frame clip in vitualdub changed the pixel format as I did it on my laptop and did not have any other encoders installed.

I apologize for the inconsistent encodings and pixel formats. I will redo the test ensuring consistent pixel formats, framerates, and encoders this evening. I will also try both of the outputs on the JVC to see if there is a difference and post the results.

Those sound like the right cables.

Here's what I'm seeing. I boosted sharpness and contrast of "captured_pattern.avi" to bring out the crosstalk pattern, and resized 200%:

Attachment 16382

That's not how you test your setup, Introducing DV into this will defeat the purpose of the entire test. DV is lossy in chroma, not to mention that your deck may have some issues in the DV section due to age as I can see clearly an advance in chroma timing of about 10 pixels, By the time the signal gets encoded to DV then decoded from DV and converted back to analog there is nothing left to rely on, Also the TBC-1000 have no effect on picture quality for this test, so it is not needed for this purpose.

Setting up a rig for analog recording is not a straight forward task, If you have a desktop you can get one of those PCIe video cards that has the S-Video output or get a media player with analog output, Another alternative is to get a HDMI to S-Video converter, But why going through all this when you can just capture a pre-recorded tape and post a sample.

Yes, the checkerboarding is very obvious in the image you posted. Obviously there is a chroma shift, but what would be causing the color bleeding to the left of all the color bars? Would this also be an effect of the suspected cross-talk?

I just got home so I will re-run the tests and report back shortly.

I agree with all of the points you have made. However, respectfully, I don't agree with the conclusion that there is nothing to gain from this test. Yes, the quality of the source is degraded after the DV conversions, but there is an obvious difference in resolution/quality on the output of my display vs the captured footage. Both my display, and capture card are connected to the S-Video outputs of the JVC VCR. So, any reduction in quality/chroma issues caused by the DV conversion would have already happened and should show up on both the monitor and the captured video. This is not the case which tells me there must be something not right with my capture card/capture machine.

I understand that the TBC is not needed. I included it to verify that is has no effect on the picture quality. If it had issues that were degrading the signal, it would show up on the captured video with the TBC, but not be present when I bypassed it.

This is a very good point. I'm not sure why I didn't think of that. I know I have a graphics card with an S-Video output somewhere, I will try and find it.

Because all of the tapes that I have are old and very poor quality. If I have a terrible quality source, how can anyone tell what issues are present on the original tape, and which issues were caused by my capture setup? Wouldn't a high quality, known source be a better signal to test with?

Yes, but there could be a problem with the recording quality of the VCR itself, That's why test tapes were made back in the day using high quality calibrated machines, But such tapes are rare to find, the closest you can get is a per-recorded tape as I mentioned above, a.k.a commercial tape that was recorded by a process called tape printing (we have a thread about it here), You may not have access to such tape but you can always get one locally or online. Then you compare the output from the VCR and from the capture card output if it has analog monitor out, if not playback the captured file.

Make up your colour bars test clip, burn it to DVD, then record that onto tape with your VCR. That should give you a pretty good quality tape. I did that with a wheel/circle test recently and it worked a treat.

Here is a chance to get a test tape.

This all makes sense. I do have several commercially produced tapes. However, they all have macrovision which would require the TBC to be inline, adding another potential source of issues into the chain.


Unfortunately, they will not ship to Canada.


This is an idea that I was considering. I will have to see if my dvd player accepts DVD-RW media as that is all I have on hand anymore. Wouldn't direct S-Video out from a PC achieve the same result?


I found my PCI graphics card with an S-Video out connection. I will get it installed in a PC and try capturing a clip directly from the graphics card output and see what kind of quality I get.

In the meantime, here are the results from my latest DV test. I left the TBC out for the whole test to eliminate possible sources of issues. Attached is an image of the internals of the S-Video cables I am using. Two coax cables with extra shielding around the entire thing.

I have regenerated the source DV clip (dv_test_pattern.avi).

I then captured a clip with huffYUV (display_output.avi).

I then unplugged the S-video cable from the capture card and plugged it into my monitor. Output is shown in (display_output.avi) and (display_output_closeup.avi).

To me, the results look exactly the same which I expected. There is no checkerboarding at all on the display. There is also much more resolution in the small lines, just like before.

I will post the results of my direct S-Video from PC, and output from DVD player tests shortly.

This is the part I don't understand either - everything else aside, if what appears on a display is different than what is being captured (using the exact same svideo connector/cable) then clearly there is an issue with the capture process/hardware right?

Are you first making a VHS/SVHS recording of the "DV broadcast" or are you capturing as pass-through? My understanding was that you were sending the DV file you created from your PC to your VCR and capturing the video in real time with no VHS recording taking place.

Yes, it doesn't seem possible to get "display_output_closeup.avi" resolution out of standard VHS. Maybe Super-VHS. Or I wonder if the LCD monitor itself is doing some kind of filtering to improve apparent resolution.

You understood correctly. I am just using the JVC as a passthrough, there is no tape in the VCR. I actually have the source video paused on the PC, so the JVC is just passing a single frame from the DV input to the S-Video outputs.


I agree that you probably cannot get that amount of resolution from standard VHS. However, since I am not recording to a tape of any kind, I feel like I should be able to get a full resolution capture of my test signal. I am also wondering if the monitor is sharpening the image.

I did notice one thing. On the original test pattern, there are 3 "groups" of small lines to the right of the center of the image, with the lines getting about twice as close to each other in each group. In my captures, all detail is lost in the first set of lines. The second set of lines is visible, but spaced out about twice as far as they should be.

I could be wrong, but this looks to me like an interference pattern. Not electrical interference, but interference of waves in physics. Sort of like there is a resolution mismatch somewhere. Here https://forum.videohelp.com/threads/...solution/page2 it is suggested that the native resolution of ATI AIW cards is 704x480. I wonder if I tried capturing the video at 704x480 if some more detail would come through. If the card is actually capturing at 704x480 and upscaling, perhaps the loss of detail in the fine lines is an artifact of the upscaling.

I will try this and and the direct S-Video output from PC test this evening after work. I am probably being picky at this point, but would like to figure out the cause for the discrepancies.

In theory it can since you don't need to compress the video to dvd format, though video cards with s-video/composite out often don''t output pixel accurate images by default and may add underscan and scale the image so it tends to be much less hassle to just use a dvd and a dvd player.

If possible I would check with the s-video output of a dvd-player or similar device to cap card to see if you get a similar issue. S-video won't look 100% clean on a high res test image source since there is some limit to the bandwidth of the format but it should be relatively close to the source.

That reminds me. In order to capture all possible detail, Nyquist says you need to capture at 2x or more of the source resolution, which means 1440 pixels across for a 720x480 source. The LCD monitor is probably capturing at a higher sample rate than 720 pixels across and that's why it looks better.