VideoHelp Forum

Try DVDFab and download streaming video, copy, convert or make Blu-rays,DVDs! Download free trial !
+ Reply to Thread
Results 1 to 10 of 10
Thread
  1. Member
    Join Date
    Apr 2007
    Location
    United States
    Search Comp PM
    First, here's a link to the problem video http://sciencezero.4hv.org/tslom.htm Series 2, Episode One, The Car. Series downloads authorized by Tim Hunkin, creator and star of "The Secret Life of Machines".

    The problem it has is about 2/3 of the way through, at the start of the segment where the car body is being straightened, it seems a field got lost and from there on the field order is backwards.

    Since the video was rendered progressive scan there's no easy fix by simply clipping the video in two, changing the field order of the second part then joining the parts.

    Applying a deinterlace filter just blurs the artifacts and doesn't fix the juddering from the bad field order. Someone tried that and posted the full episode on YouTube (how do they post a video longer than 10 minutes?) and the result is terrible looking.

    How I'd like to fix it is to create fake fields by masking every other line, starting at the top line then at the second line, then re-order them and re-render to progressive scan Xvid.

    Is there a way to do that without a huge amount of fiddling and twiddling?

    I know it could be done by importing the video into Photoshop, creating masks to make every other line transparent, piecing the correct parts together, flattening then saving every frame. Next step would be importing that frame sequence into an editor and rendering as video. Should be able to automate it with scripting.

    The other part of the video would have to be converted to individual frames to import into the editor along with the fixed part.

    Has someone already worked out how to do this? Of course the faux field masks would have to be made custom for the video, but the scripting should work ready made.

    If there's something like a VirtualDub plugin or other software with a function to do this, that'd be even better.

    Of course I'll share the fixed version. Every other copy of that episode on the web has the same problem. There must have been just one capture made from the master tape and every digital copy has inherited the error.

    I remember when this show was on Discovery, that episode didn't have the field order problem. I wish I'd taped it, then I could do my own capture and conversion.

    This shows how important it is to check your digital output for problems. Apparently whomever captured these didn't. fortunately it's only "The Car", the other episodes are all error free.

    Thanks much in advance. Other fans of the show will likely be happy to get this episode fixed.
    Quote Quote  
  2. Member 2Bdecided's Avatar
    Join Date
    Nov 2007
    Location
    United Kingdom
    Search Comp PM
    544x352 29.97p

    It's not just a simple field order issue back up the line (which would be bad enough to solve after downsampling to 352-lines!!!) - the whole video is jumping, and colours and luma (separately) are repeated from previous fields/frames - not just the adjacent ones though.

    Quite a challenge - I have no idea! Just wanted to post this so other people might have a better idea of the challenge.

    Cheers,
    David.
    Quote Quote  
  3. Yes, the video was resized so the two fields have been blended together. There's no fixing it. You would need to go back to the original cap to have any hope. About all you could do is blur the frame on the vertical axis until the comb lines have disappeared. Then apply a little sharpening along the vertical axis. In the end, you may get rid of the comb artifacts but the picture will be blurry, look like a double exposure when motions are large, and the other defects will still be there.

    original on left, processed on right:
    Click image for larger version

Name:	example.jpg
Views:	131
Size:	38.5 KB
ID:	12547

    Code:
    BicubicResize(544, 118)
    Nnedi3_rpow2(4)
    BicubicResize(544,352)
    Sharpen(0, 1.0)
    You could do better if you're willing to put in a lot of time -- by replacing individual frames with either copies of nearby frames, or motion interpolated frames. But that will require going through the entire video, identifying which frames to replace, then adding commands to replace them.

    Motion interpolated frames:

    Click image for larger version

Name:	pic0000.jpg
Views:	123
Size:	89.2 KB
ID:	12548
    Click image for larger version

Name:	pic0001.jpg
Views:	182
Size:	89.9 KB
ID:	12549
    Click image for larger version

Name:	pic0002.jpg
Views:	124
Size:	89.3 KB
ID:	12550
    Click image for larger version

Name:	pic0003.jpg
Views:	144
Size:	89.4 KB
ID:	12551
    Click image for larger version

Name:	pic0004.jpg
Views:	140
Size:	90.1 KB
ID:	12552
    Click image for larger version

Name:	pic0005.jpg
Views:	152
Size:	91.2 KB
ID:	12553
    Click image for larger version

Name:	pic0006.jpg
Views:	116
Size:	91.3 KB
ID:	12554
    Click image for larger version

Name:	pic0007.jpg
Views:	131
Size:	90.0 KB
ID:	12555

    The six frames in the middle were interpolated from the first and last frame. That was a case where the motion interpolation worked well. But it doesn't always work that well. A case where it didn't work well:

    Click image for larger version

Name:	poor.jpg
Views:	140
Size:	39.7 KB
ID:	12556
    Last edited by jagabo; 25th May 2012 at 10:27.
    Quote Quote  
  4. Member
    Join Date
    Apr 2007
    Location
    United States
    Search Comp PM
    Is there a way to do a weave that doesn't do any interpolation, just slices the frames into lines and the lines to swap between frames can be chosen?

    It'd be like taking a series of still images that have been cut into lots of strips then strips from adjacent images have been mixed up - the goal is to un-do the mixing.

    I grabbed a couple of frames and have been fiddling around with various stripe masks. Looks like it'd take four masks, each with a 1 pixel black line every 4th line and each mask offset down 1 pixel. Choose which of the four that makes the frame look like it all belongs together then select the other masks in the other frame. Flatten, select the non-black lines from both frames and paste both to a new image. Get the right lines selected and the new frame has little or no interlace artifacting.

    Create the masks, figure out which pixel lines to slide which way and how far (immediately adjacent frames or two frames away?), then figure out how to script the slicing and sliding in Photoshop. Should be possible but I'm not a Photoshop wizard. Would be even easier to have a plugin for a video editing program or a standalone program. Don't do any interpolation or blending or blurring, just render out every frame then swap whole horizontal lines of pixels back and forth according to user selection.

    It would need plenty of adjustments because of differences in each video. How many line grids, which lines to shift which way and from how far away? Without doing any pixel to pixel or line to line interpolating I think this could be a pretty fast process.

    Since it's a large section that's screwed up it shouldn't matter if the pixel line interleaving is applied to the entire segment, might even sharpen up frames with little motion that don't show much or any of the problem.

    Would be better if I could get a copy of the 640x480 Quicktime version but those have been offline for years. What I linked to is the best currently available.

    I could see such a tool being useful in professional video work where master tapes have problems like this that can't be corrected during capture, and for cases like this where all someone has is a digital copy and no longer has the original analog master. Another tool for the video wizard's kit.
    Quote Quote  
  5. Originally Posted by bizzybody View Post
    Is there a way to do a weave that doesn't do any interpolation, just slices the frames into lines and the lines to swap between frames can be chosen?
    Of course. But that video doesn't have two separate fields that you can seprate and recombine or deinterlace. The two fields have been partially blurred together. There's no way to separate them now. I have two numbers. The average of the two is 32. What are my two numbers? There 's no way for you to know. If you had completely still frames you might be able to restore the original frame by the technique you describe but it won't work with moving images. It's not as simple as every 4th line.

    Here's every 4th scanline as separate images (each 1/4 as tall). The original frame is on the left, the four stacked frames on the right. The top one is scan lines 0,4,8,12..., the next 1,5,9,13..., the next 2,6,10,14..., and the bottom 3,7,11,15... As you can see, none of them is clear of comb artifacts.

    Click image for larger version

Name:	four.jpg
Views:	137
Size:	54.1 KB
ID:	12566

    Every 3rd line is a little closer but still no cigar.

    In addition to that, you the video was originally PAL and it was converted to NTSC with field blending. So there are many fields that are a blend of two PAL fields or frames. Then you have the time base problems which caused one field to jump up or sideways by a few dozen lines/pixels sporadically. And finally you have chroma blending problems. It's pretty hopeless.
    Last edited by jagabo; 25th May 2012 at 21:28.
    Quote Quote  
  6. Can't you just order the DVD or is that messed up too ?
    Quote Quote  
  7. Member
    Join Date
    Apr 2007
    Location
    United States
    Search Comp PM
    I e-mailed Tim Hunkin. He replied he's going to see about sending a good DVD made from his original tapes to the guy who did the videos on the Science Zero website.

    Those videos were made from the US version which were all around 25 minutes. The original UK version was over 27 minutes per episode. If that just gets a good copy of the US version of The Car to download, that'll be great. If they make the longer UK versions available to download, that'll be super great.
    Quote Quote  
  8. I've never seen or heard of the show before - it looks like Mythbusters 80's style! Pretty cool

    Would be better if I could get a copy of the 640x480 Quicktime version but those have been offline for years. What I linked to is the best currently available.
    Seems to be up ?
    http://www.exploratorium.edu/ronh/SLOM/0202-The_Car-big.html
    Quote Quote  
  9. Yes, it looks like an interesting show. I'm downloading the 640x480 clip from the Exploratorium web site. It looks like it's a little cleaner. I'll post here if I find anything interesting.

    <edit>

    The larger frame video at the Exporatorium didn't have the screwed up comb artifacts but it was blend deinterlaced and still had all the other problems. That may be easier to fix but the video is so bitrate starved all the motion areas are a blurry mess. Not worth bothering with.

    </edit>
    Last edited by jagabo; 25th May 2012 at 23:16.
    Quote Quote  
  10. Member
    Join Date
    Apr 2007
    Location
    United States
    Search Comp PM
    Originally Posted by poisondeathray View Post
    I've never seen or heard of the show before - it looks like Mythbusters 80's style! Pretty cool

    Would be better if I could get a copy of the 640x480 Quicktime version but those have been offline for years. What I linked to is the best currently available.
    Seems to be up ?
    http://www.exploratorium.edu/ronh/SLOM/0202-The_Car-big.html
    I always get a 404 on the Quicktime versions. Are you in the USA? I am.

    If you have kids in their early teens, this is a great show for them. Ought to spark some interest in how things work. It's also great for anyone who wants to know how things work.

    One exception though, in the Television episode he didn't name the right person as the inventor of electronic television. That was Philo Farnsworth. Everyone else was dinking around with partly mechanical systems that were very poor performing. RCA tried to steal Farnsworth's invention but he was the first to foil them. Up til then RCA's methods were to first offer an inventor a large amount of money to buy their idea outright, no royalties. Then they'd either do nothing with it (to prevent competition to something they'd already put a lot of money into) or produce it when they figured there was big money in it. If the inventor couldn't be bought, RCA would file their own patent then sue the real inventor*. Whichever way it went down, RCA always got what they wanted, until Philo Farnsworth who just wouldn't go away and let the big guys take television from him. RCA did manage to drag it out until Farnsworth's patent was almost expired so they didn't have to pay royalties for very long.

    *A practice still used by some companies to this day.
    Quote Quote  



Similar Threads