Restoring vs. distorting, audio and video?
Hi everyone,
I've been lurking for a while and reading a lot. I've been hanging out at videohelp for a while too. I got an account here yesterday and downloaded the Lord Smurf VDUB. There is some good stuff in there. Sorry long first post............... Anyhow, some background: I have 12 plus years experience as an audio restoration hobbyist. That is, taking analog audio from cassette, vinyl, VHS, and Beta and cleaning it up for digital archiving. I am in the process of restoring family videos from VHS, VHS-C and mini-DV. I have already captured all of my footage in DV and HuffyUV AVI (VDUB 720x480). Now, in my school of audio archiving some concepts carry over to video but some do not. I believe in creating a final product that represents the original as faithfully as possible. In audio archiving, noise reduction is the work of the devil. Audio enthusiasts will come after you with pitchforks for using it. NR kills detail in audio and I am learning through experimenting it does the same for video. In video it creates splotchy patches and blurs out detail. Like in audio, in most cases, I'd rather deal with the noise and retain the detail. Some enlightenment on how to properly use video NR is welcome, or I may just skip it all together. Fair play in audio restoration is the use of EQs and notch filters, de-clicking/clipping (done properly) and the occasional compressor/limiter. These tools can be used to shape the sound to make it properly represent the original event. I am light handed when it comes to these tools. I often reduce undesired frequencies rather than boosting desired frequencies (boosting being an armature mistake). How this translates to video I'm not sure and I am looking for guidance. When I am correcting color is it better to reduce the undesired color like reducing unwanted frequencies in audio? There are obviously more "knobs" to tweak in video and I find myself getting carried away. After hours of tweaking color, contrast, and white balance on a clip I get lost. I think it looks better but maybe it's too artificial and I am punching settings too high. How do I benchmark and determine if I am helping the video rather than just repainting each frame with a digital facsimile? I guess I am looking for advise on the best do's and don'ts using the LS compiled VDUB toolbox. Specifically for VHS. My delivery format will me MPEG2 as I have found it to be the friendliest codec for VHS (what I use to encode will be another thread :) ) THANKS! |
To restore is to return something to it's original state; in this case it would be to view a VHS as if it just came out of the camera.
Without the original camera to compare it too, it's impossible to know to restore the colors perfectly. Why do we restore? My guess is to preserve a feeling of authenticity, and details of techniques used at the time. You can also enhance. In this era of HDTV, we don't expect noise and blur. The choice is up to you. You can also fudge the definition of restore to say, you want it to look as realistic as possible; as if you could go there again with modern equipment. That means making skin tones look proper even if "authenticly" the camera always had bad white balance. Theoratically, except for TBC and repairing dropouts, you wouldn't do anything to a VHS, as it probably doesn't look that much different than originally. VHS doesn't fade colors over time AFAIK. Even TBC is cheating; unless it was part of the camera playback, it's not authentic. The equivalent of EQ is sharpness. There are some light methods of denoising that barely degrade the picture. Clicks would be dropouts or white lines. As far as boosting vs damping, the order of operations will affect numerical accuracy, and dealing with 8bit video, there's not much accuracy to go around. Another point is that all your work on color grading will be for nothing unless you are using a calibrated monitor. When you've made it look perfect on your equipment, you may be disappointed to discover that it doesn't look the same on another TV. Most TV's these days vary quite a bit in their look and it takes a lot of work to calibrate everything. I'll let the experts address the rest :) |
Enhance it the word
Great point about monitor calibration. Any suggestions on a reasonably priced calibration package? Seeing that my other hobby is digital photography this would be handy. I don't want to derail the topic though.
I like your point about restore verses enhancement. I think those terms get minced in these forums. To me a restore would be getting the tape to play and nothing else. Kind of like fixing up an old car with only stock parts. Where enhancing the video would be like putting some chrome wheels and a flame paint job on the car or variations thereof. To that, I suppose I have already restored the video. I would like to enhance it tastefully by possibly removing noise and adjusting color to something that looks natural. I guess I want it to look better without looking like I did something to it, if that makes sense (no overworking or smudginess). It seems like some people want to take their VHS and make it look like it was never VHS, this is not my intent. In fact I don't mind it looking like VHS, it dates the footage perfectly. However, some of the footage is full of chroma noise and bad white balance. Some has blown out exposure from 2nd and 3rd gen SLP copies, just awful looking but I'm sure it can be spruced up a bit. I thought I'd ask you all for advise before I go bushwhacking away on filters. |
Yep, I'm replying backward to this one. jmac first, then magilla (the OP).
Here goes... Quote:
This is, of course, a definition limited to video -- not films, fine art paintings, etc -- just video. The "original state" of video is very often not all that good, and in many cases restoration is being sought to improve it greater than it's original condition ever was (or in some cases, could ever have been, given limits of how/when it was shot or recorded). Quote:
Same for VCRs that record off TV. The goal would to be to make it look as good as the original broadcast signal it was recording, not the recording itself which is for whatever reason degraded. Quote:
For example, tape hiss and consumer audio cassettes. The original tape was a dull sound, and added noise. Restoring this tape would remove the hiss, and calibrate the audio back to normal levels and EQ -- maybe even removing audio distortion (red-lining of values, which can cause crackles/etc). Quote:
Archiving "as is" or "true to the format" is also not restoration -- there is a distinct difference in restored and archived, and a librarian at an archival facility can probably (hopefully) give a good insight on that. Quote:
Quote:
Quote:
Quote:
Consider a nice B&W photo. -- Preserving it means to retain all the original sepia tones caused by the chemical development process, possibly even the various imperfections (hairs, grains, dust/dirt specs and spots, etc). -- Restoring it involves removing the hair/yuck, cracks, and balancing the sepia. Sometimes completely removal of sepia is desired, as that can be a purely aging issue, and not how the original looked. (I have prints that are starting to approach 20 years old, and the day those came from the wash bath, I assure you the images were black and gray and white -- and that's it. Now they have red/yellow tints.) Quote:
Quote:
Quote:
Quote:
Failure to pre-filter will result in blocky/ugly video at most "format" bitrates (DVD, Blu-ray, web streams) -- and this is where audio and video differ. You could leave all the noise in an audio file, but you can't do that for video. You will create FURTHER damage to the signal clarity by NOT filtering any. This is why I got so involved in filtering and restoration in 2001, because I saw how badly DVDs would look when no effort was put forth. It's worse on homemade sources. (Note that I had already been doing restore/filter work on tapes, back in the 1990s, but it was more by choice then. With DVD, it became necessity.) Quote:
We're really not too different on our approach to filtering. Reduction of undesired portion of the audio (frequency carving, as I call it) is how most of my audio filter work is done. Sometimes cancellations are possible. Boosting is only acceptable when the original so far far screwed up that it's the only way to pull out what little good audio is left -- when the source is more noise than audio. I run into this quite a bit on OTR (old time radio shows) that were recorded with wire, cones, vinyl and others. Quote:
Quote:
Quote:
|
Quote:
Quote:
Restore - improve it without changing the actual content (the noise added by the recording medium is not content) Enhance - to alter the content In photography terms: Archive/Preserve - print/scan, no corrections (this is a "picture", not a photo) -- like a JPEG from the camera Restore - process photo in darkroom, Photoshop Enhance - Photochop, makes an "illustration" (no longer a photo) -- HDR goes here, too. Overused gimmick. NOTE: You can archive restored versions. There is some contention here amongst galleries and libraries, and I think it's stubborn and bullheaded for "uncorrected" to be the mantra espoused by some places. Archive them both. Some people want the best version available to see/hear, not the one that's falling apart. Quote:
To me, it's like keeping a broken leg instead of using a cast. Fix it. |
2 Attachment(s)
sample1
This is badly shot 2nd or 3rd generation SLP. The color is blown out and there is plenty of chroma noise. This is one of the worst clips I have. sample2 This one is pretty bad too. Most of my clips are in pretty good shape. If I can get a grip on how to clean the really bad ones I think I'll be headed down the right path. Thanks for any guidance! |
Hi,
Interesting discussion! I didn't know that the restoration definition differed in the field of video; for the record I did check on the definition before I replied. Here is what I found: Quote:
Obviously in film there is some controversy; do colorized versions count as restorations, even if the original director supervised it, even if he meant to shoot it in color? Does "Han Shot first" qualify? I can see leeway for the argument that re-creating the original broadcast regardless of the method used to record it would be a restoration. As far as restoration vs forensic, it seems by your own admission, the purpose of forensics is to answer a question about a particular fact or detail regardless of the means necessary to obtain it; in your very own example, to see the structure of a face regardless of it's skin tone, where restoration would include seeing the skin tone as faithfully as possible. In terms of film there is another important distinction; there is considered to be an original artistic intent relevant to the processes in use at the time; film grain can be used *on purpose* to achieve a specific look. To "restore" such a film by removing the grain would be destoying the artistic integrity of the film. Though I accept as respect your opinion, I'm still not convinced there isn't a *controversy* over the details of the definition. Anyhow, my understanding of the original poster is that he is seeking "natural enhancement" (no pun intended :), that is, improving quality without introducing a subjective perception of un-naturalness. Certainly, this everyone's goal. I would improve the definition of archive/conserve/preserve as "prevent the occurence of age-related detoriation; to maintain extant quality without change", restore might be "to remove medium related defects which weren't present in the original", and enhance as "to alter the original form to add qualities not present in the original". These definitions might overlap. EQ vs Sharpness: actually these are exactly the same. If you took a video signal and put it into a WAV file, then put EQ on it, then converted back to a video, you would have sharpness. I've experimented with this; treating video as audio to see what would happen. For example, the simple formula for average (pixel N+1 added to pixel N divided by 2) has an exact translation into the frequency domain as a low pass filter; see the blue line in this diagram: http://ptolemy.eecs.berkeley.edu/eec...esponseRA.html A sharpness filter might subtract the original samples from the low-passed samples and be the same graph "upside down"/mirrored (?), or high pass filter. VHS colors don't fade - from my technical perspective, the reason would be that color is stored as an FM signal. In this modulation, a certain frequency represents a voltage value for the color signal. See http://books.google.com/books?id=NXE...%20vhs&f=false A frequency of 629KHz represents 0, a higher frequency represents a higher voltage, respective for a lower voltage. The frequency of the voltage going up and down (and the frequency of the 629KHz) gives the tint, and the amplitude gives the saturation. A wild swing from (say) 600-700Khz means a deeply saturated color. This is an indirect way of storing color. If the tape "fades", the carrier (629KHz) is less. but as long as it can be detected, the color it represents doesn't change at all. I'm not sure how well I explained that but I won't write an engineering course here. I'm not sure what color a dropout will turn out to be; but some specifc color for the duration of that dropout, perhaps greyscale. I agree with your example of the photo; to me that's just restoration. You talk alot about compression artefacts, but he was referring to VHS, VHS-C and some mini-DV. I think he talks about NR being too unnatural in terms of cartoonish, or blurry. That's definitely a tradeoff in all denoising methods I've seen. Interesting link, upsizing methods: http://www.general-cathexis.com/inte...ion/index.html Quote:
http://forum.doom9.org/showthread.php?t=154719 I tend to ignore audio and focus of video; but it would be interesting to hear what you guys have done for audio restoration/enhancement whatever. One neat program I found is: http://www.winamp.com/plugin/tape-restore-live/154246 - Software Dolby B remover Approximates a Dolby B decoder in software. If needed, tape bias settings can be fixed first. This is especially useful if the sound of tapes has become dull due to tape wear and age. Also, sound card noise is deminished by the remover. - Automatic AZIMUTH correction If the tape head is not set exactly equal to the setting during recording, playback in mono or through a surround system will cause very ugly artifacts. This filter corrects for this on the fly. - FM Stereo hiss removal Removes the FM stereo hiss without damaging the real sound. (Not only useful for tapes!) ps I have an experiment where I have the original digital file which was recorded to tape and then the tape version; I onced owned this cassette deck - http://www.generalmanual.com/Audio/P...sette-Deck.htm Results - cassettes degrade severely! The high end was really lost, the Dolby tracking was lost, the tape sounded just horrible and good riddance to cassettes. I'd rather have open reel for an analog recording. |
Quote:
Quote:
Indeed, there are arguments to be made for color vs B&W, pre-CGI vs post-CGI versioning (example: Star Wars), and both sides have valid points. What Lucas did NOT do, however, was leave the film deteriorated and transfer as-is, nor did he leave in obvious mistakes from the first edition (matte outlines around X-wings). Those aspects do qualify as restoration. Indeed, there have been some fans that actually restored it better than Lucas did, by hand-painting thousands of frames, using every available source as the starting point (including some of the restore work done by Lucas, but excluding his alterations). ... magillagorilla, I'll be sure to look at your samples here in the next couple of days. :) |
1 Attachment(s)
Some movies have grain on purpose like the opening scene of Casino Royale. I think that example speaks for itself; why else would a modern film be shot in B/W? On the other hand, you have a point - you didn't have a choice to avoid grain completely. Also I have a great example of a bad remaster, dunno if you've heard of it but Krafterwerk Home Computer remastered was horrible. Perfect example of the unnatural effect. The audio has a "drowned" effect, to me it sounds like mistracking Dbx, or a compander set wrong. It cuts off the natural attack and decay of sounds. I'd rather the noisy version.
Another thing, the so hyped Beatles remasters, don't sound any different to me, I can't see what the fuss is about. Anyhow. I looked at the sample. A lot can be fixed with white balance correction. I can correct the color stripes, in fact I did this already for another poster. The sample2 has some weird color lines that I'll have to think about. Give me some time. -- merged -- Here ya go. The colored lines are almost all gone, the big purple haze on the left is gone, the color is somewhat corrected. The comets are gone. You should try my favorite technique, capture the same section 3 times and post. |
Quote:
Nice, was that just white balance or did you punch down the blue on a color filter as well? |
I tried a number of combinations, autowhitebalance seemed to work best.
Your other sample is a bit harder, for example I can't use autowhite balance because it leaves a correction tinge on the blowouts. The flashing colored lines at top might be harder too. |
Any hints on what specific filters and settings were used. I know this site offers a service and I don't expect anyone to give away the farm but the LS VDUB has a lot of filters. Just looking for pointers on which ones may best suit my clips. Maybe some hints on settings so I don't "over do" it.
Is there any method other than eyeballing it which I should consider? |
Actually I used Avisynth. I don't have time to completely clean it up, it's just a start. The scripting is really messy too :)
Code:
vid2=MPEG2Source("G:\project001a\sampleproblem\sample2.d2v", cpu=0) I hope you have something better than the mpg version, it was really blocky. |
avisynth
Thanks, I'll try it out. I don't really know how to use AVIsynth but I'm no stranger to codeing. It does not look much more difficult then editing an INI or a JAVA .config file. I just don't know the toolset or parameters at all. I guess I could read the manual.
My source is HuffyUV played back from a Pany AG-1980P. It isn't going to cap much better unless someone has a Datavideo TBC they are selling for $40. I doubt I can afford any more hardware. The huffy files are not blocky at all. The sample was coded on a crapy 1pass low bitrate. Hopefully when I do MPEG2 2pass VBR 6000-9000mbps it will look good. Or I may go with h.264, I havent decided. |
Ah, that explains it.
What I meant was, there's a technique where you cap the same thing 3 times in a row. This seems redundant, but a lot of types of noise only appear the same place once. I'm also curious if that's the case with your particular problems. As far as the script, it will take you so far, but you'd have to learn the scripting to make any improvements, that is, unless you pay these guys to do what they do. I did write a quick tutorial elsewhere here. In your case, you should write vid2=avisource("") for your original file. assumetff can be removed (it only applies to mpeg2). The d2v file you don't have, I created it with a program called dgindex, but that doesn't matter now. Understanding my script is going to be a little hard as it's messy. If you want to code with it, get a program called avspmod. It's a nice, code completion ide and video preview. Now in general terms, what I found is one of your heads is bad (I believe?), the comets are little white streaks that appear, they only appeared in one particular field. The color lines were also worse in that field. So I just used the good field. Also I found that the color lines were just in one color channel, so I heavily denoised that channel. As for the purple haze, that is transparent, so I just color-corrected that section and it comes out fine. Ironically you could also use a good transparent logo remover to do an even better job. |
hm, I hope I can get some of this done in VDUB since the filters are GUI. I'll try the logo remover. Also, I concure with your theory of a bad head. The bad head was, however, on X generation and not the final playback deck. I have cleaner tapes with no comets on playback using the same deck.
|
Actually, you can use the script itself as input, and then add vdub filters on top of that, so it's no problem - just keep going!
|
sample1 is too far gone. The video signal appears to be pure luma with blue chroma and no red (in other words, just the YCb -- no Cr -- as expected in YCrCb signal). The image is incomplete, and there's really no way to fix it. There's also therefore no reference point upon with to restore any kind of white balance. At best, we could shift colors, but recreation/restoration is essentially impossible.
Footage this badly destroyed should be converted to grayscale (B&W video). sample2 is not quite a badly damaged, but still appears to be missing some portion of the full signal. At best guess, the original camera was damaged when this was shot, because tapes don't "fade" color-wise. The only other option is this is a copy of a copy of a copy (etc etc etc), without a TBC, at such an nth generation that it's impossible to recover a proper signal. jmac gave it a good try, but essentially just altered the remaining signal colors to a sepia-shaded tone. Again, conversion to B&W would be the ideal solution. A B&W signal is more acceptable and watchable than a severely color-shifted version. (Feel free to archive or "preserve" the original, but create the B&W for pleasant viewing conditions.) Some things can't be restored, unfortunately. This is a good example of that. Wish I had better news for you. :( |
Quote:
|
1 Attachment(s)
Well I have better news :)
The colored lines at the top are fixed, this is easy and automatic. The color haze at the left can be fixed; there's still some left only because I didn't finish tweaking it. The comets are gone, that's standard. The colors are corrected; except for that yellow on highlights - I'll get to that in a bit. It's been denoised and deinterlaced as well. Never say never :) |
Site design, images and content © 2002-2024 The Digital FAQ, www.digitalFAQ.com
Forum Software by vBulletin · Copyright © 2024 Jelsoft Enterprises Ltd.