The only difference in the two methods is being able to change compressors or color space, etc.
I can't find the previous post where you stated you tried to download "algorithm" ? ? I think you meant Lagarith. Get it here:
http://lags.leetcode.net/codec.html . Many people don't use anything else.
To address the bad frames and shifting: it's not always your fault. It can easily be the source.
The tape may be damaged, but in any case is not moving smoothly and is giving the TBC and AVT a rough time. A pro sync machine at 10 times the price would have 10 times the memory cache and processing power, but with pro-sumer gear you have to try to help your hardware a little. Often repacking a tape will offer smoother feed off the casette's spool and more consistent flow across the heads.
Repacking means rewinding the tape a time or two, without playing and without pause. Examine the tape cassette from the top. If you see that the tape winding on the left or right side has a lot of bumps, ridges, and uneven winding, it can cause jerky tape movement. Insert the tape into the VCR, but don't play it. Fast-rewind all the way to the end, without pause. Then fast-rewind all the way to the beginning, without pause. You might have to repeat the process. The idea is to rewind the tape more smoothly onto the right-hand feed spool.
VirtualDub reports that the video is slightly off-speed at 29.971 fps (average, or just heaser info). An AVisynth plugin that I used reported that the frame rate was inconsistent. Disturbances in tape motion can occasionally cause inconsistencies in a few frames. An Avisynth function (AssumeFPS) can help with that.
First, a closer look into what the video is doing. Open the clip
"1986_retail_with_AVT_and_panny_TBC.avi" directly in
VirtualDub using "
File..."->"
Open video file...". You won't be using filters yet. As with other editors or PC players, YUY2 video will be converted to RGB, but only for viewing. The file itself won't be affected because viewing is all you'll do for now. We need to look at specific frames, so use the frame advance ">>" and reverse "<<" buttons at the lower left of VirtualDub's window.
The video is interlaced. Keep it that way for the time being. Use the VirtualDub ">>" advance button to view one frame at a time. You can move at a more normal rate by holding down the "->" directional key on your keyboard. Move forward, and stop at frame
92.
Frame 92 has a bad dropout across the top. This could be damage, or it could be a blob of runaway dust. It may appear again in this or similar location if recaptured, but likely it will always be in the same frame. Avisynth can help with that, too.
Move to frame
394. You can move back and forth between frame 393 and 395 to see what's happening when the image appears to "shift". Actually it's only the bottom half of the frame that doesn't move properly. The bottom half shifts left before the top half does. The bottom half is also blurred. Repacking the tape and recapturing might prevent this. If not, Avisynth can offer some help. It's also possible that the original film has that defect.
Frame
672. This frame shifts upward and to the right. The upward nudge is called frame hop or "projector hop". Frame
673 moves downward and to the right, and is blurred. Frame
674 is normal. Possibly the original itself has some camera movement or object shift to begin with -- you'll often see this, especially in promotional films. I'd guess it looks like a production problem. Likely you can't avoid the shift altogether, but you can smooth it out a bit with Avisynth.
All sorts of things go astray in the first shot of the clip. Look at the landscape and foreground as the camera zooms in. Literally, objects and planes are rolling and shifting, lines twitter or act broken. This could be a fault in the original (possible), or poor tracking from tape damage. Again, I'd suggest again that you repack these tapes if the windings look sloppy. It often makes a big difference, and it helps the player do what it has to do.
You wouldn't want to fix any of these defects while the clip is interlaced. Now, let's look at this video
field by field rather than frame by frame. Often you'll see a frame problem, but the problem exists in only one field of that frame.
I used Avisynth to perform the following, but for now you can use VirtualDub to see similar results:
Move the video back to its beginning, at frame 0. Go to "
Video..."->"
filters..." and choose VirtualDub's "bob doubler." Set the field order to "Top Field First" and the deintelace method to "bob", then close that dialog to activate the bob doubler. The fist thing you might notice is that the frame count of the interlaced file began as 780 frames, but the count is now doubled to 1560. The bob filter isn't the best deinterlacer. It's rather crude and gives a soft image that often flickers. But it's handy as a quick analysis tool. The interlaced fields are now separated and resized to full-frame 640x480. Remember, now, that all "frame numbers" will be doubled. The original field 0 of frame 0 is now bobbed frame #0. The original field 1 of frame 0 is now "frame" #1. When a video is bobbed or deinterlaced, the new even frame numbers 0, 2, 4, 6, etc., were originally the even fields. The odd numbered "frames" were originally fields 1, 3, 5, 7, etc. To calculate which original interlaced frame you're viewing, divide the bobbed "frame" number by 2.
Move to bobbed frame number
184. Remember that the
left-hand window is the original video frame. The
right-hand window is the result of the bob. In the
left-hand view of the original frame, you will see the dropout flash in frames 184 and 185. On the
right-hand bob-filtered view, the dropout streak appears only in frame 184. This means that the bad streak is only in the even field of the original interlaced frame. The original odd field of that frame has no dropout. The original interlaced frame number is 184 / 2 ( = frame 92).
Just out of curiosity, move to bobbed frame
202. You'll see a white spot in the upper right sector. This white spot appears in bobbed frames 202 and 203, so it's in both fields of that frame. Moreover, the spot is in exactly the same place in both fields. If you deinterlaced this video and used a spot remover to clear that spot, the spot filter would see it in two consecutive frames. So that spot would likely stay as-is. If a disturbance doesn't move, it's often not seen as "noise". But there are ways around that, too.
Move to bobbed frame
788. This is the frame that had the shift in the lower half of the image (original frame 394). If you move between bobbed frames 788 and 789, you'll see on the right-hand side that the shift occurs only in the odd field of the original frame.
Finally, move to bobbed frame
1344 (original frame 672). This is the shifting robot. The object shifts up and to the right in frame 1344-1345. But notice that in the left-hand window, the wand in the robot's hand doesn't move. In the
right-hand window, the robot's arm moves in bobbed frames 1344 and 1345. In the left-hand "original" window, frames 1346 and 1347 are alike and look blurred. But in the right-hand window, bobbed frames 1346 and 1347 show clean movement. The same right-hand/left-hand thing occurs in the next two bobbed frames. This indicates that in order to repair this 3-frame interlaced sequence, you have to deinterlace and look at the problem as stretching across several fields -- perhaps up to 6 fields, maybe less.
The attached mpg is DVD spec. It's just an exercise, really, to see what I could get. I repaired most of the bad frames mentioned, cleaned some bad chroma bleed and shift. But more could be done. The first scene is a lesson in what results when you try to fix everything in sight: over-filtering, usually. Color balance is fantasy here, just to test effects.
Statistics: I used Avisynth for most of this. Deinterlaced with QTGMC, then resized with Spline36Resize to 720x480 and reinterlaced for encoding at the end. Several denoisers were used, including MCTemporalDenoise, Santiag, and cnr2. Most of the bad frame repair used ReplaceFramesMC, which is based on MVTools. Some of the color stuff was VirtualDub (ColorMill and gradation curves). Encoder: TMPGenc Plus 2.5.
I doubt that most users would go thru this much trouble for every video. Life is too short.
ED: Saw a VirtualDub guide similar to this post some years back in digitalfaq, but couldn't find it.
