@motleymongoose, let's get more complicated with a video that has several more problems than yours. You're bound to encounter something like this. I dug up some old samples that use most of the same VDUb filters and some of the same Avisynth plugins used in the previous posts.
In this case I had an old family video from my sister shot with a cheapo VHS camera whose autogain and autowhite made a mess. She ignores strong backlights and other stuff that ruins exposures. Here I present a case with a problem that's the opposite of yours: far too dark, grossly underexposed with crushed blacks. The original tape was damaged by a JVC later but I archived the huffyuv captures made earlier with another player. The captures are 640x480.
Most of the tape looks OK, but about 5 minutes is garbage. The samples include two edited scenes that I called "bad shot1" and "bad shot2". Here are samples from two unfiltered frames:
bad shot1
bad shot2
Most consumers think auto cameras will give them the same thing they see in the viewfinder. But they don't. Here, one can forget getting any detail out of the blown-out overhead light in the top image. Those details were destroyed during the shoot. Meanwhile the AGC obscured the man in the shot. You could just "brighten" it, but when the camera pans right the video brightened again and the woman standing next to him would be blown out of the picture. I don't know what happened in the bottom photo, but it's underexposed in soft late afternoon hazy sunlight and is too blue. During the short scene you can see color and gamma change 3 times.
One of these years I'll take up this cap again and work some more, but as of a couple of years ago these are results I got with Avisynth and VirtualDub:
An mkv of the original edited bad shots joined together is attached as home_tape_original.mp4. The second shot outdoors has bad moire and twitter in the lattice work of the white bench in the image, and looks quite "disturbed". Try to ignore the idiot at the right who was clowning for the camera. Both shots have the typical aliasing on motion that you get with consumer cameras, and object displacement as in the first two swimming pool captures thanks to the camera's shutter. Still need to work on that red chroma shift (headache). The captures are 640x480.
An mpg of the filtered shots is attached as home_tape_rework.mpg. One might get rid of some of remaining edge twitter, but not much video would remain.
As examples of advanced cleanup, the scripts I used for each shot are these:
I don't like autoplugins. They never do what you think. The levels balancing act used HDRagc. With default settings it will blow a dark video into a juiced up monster that leaps off your screen and lights up the neighborhood. It's considerably toned down here. I used it first before QTGMC to give QTGMC a break with bad levels, which can look worse when deinterlaced, and because HDRagc works in YUY2. SeparateFields() was good enough, according to HDRagc's instructions. The output is 704x480, which is valid for 4:3 DVD and SD BluRay. A 16:9 video would require 720x480.
Any comments on the script are OK with me.
@motleymongoose, at least your clips don't look as bad as these.
The following users thank sanlyn for this useful post:
archivarious (12-09-2020)
Thank you very much for all of the good information. I don't know why but, I have not been receiving forum post notifications. I thought you had forgotten about me. Boy was I wrong. It has been a hectic week. I am going to have to take a minute to digest all of the great information.
I am still struggling with the RGB stuff you were talking about. I need to learn how to read histograms. I think I understand that you are trying to squeeze everything into the middle if I may grossly over simplify. Is that correct?
You can oversimplify (tutorials do it all the time), but it should refer to the correct operations, LOL! I believe all of the color work has dealt with stretching color intensity and brightness in both directions, mostly outward, not inward into the middle. The middle of the spectrum is neutral or muted hues, or what photogs and cameramen refer to as "middle gray" in brightness, while the left-hand quarter would describe shadow areas that darken down to blacks. The color and brightness values in your two samples are already in the middle of the spectrum, with very little in the dark range and nothing above the middle, and bright reds are oversaturated. If your sample is supposed to look like a scene shot through a fogged-up camera lens, then you would leave it as is. If the "before" unaltered images shown in these posts look OK to you, and the "after" corrections all look distorted, then you should avoid color correction.
All of the "after" samples have had some denoising as well, which makes changes in their appearance and actually does change the behavior of many objects, for better or worse depending on what you're trying to clean up and how you do it.
Histograms don't fix anything. They just show you what's there. They're information, not corrections themselves.
I understand that the histogram is just information. I just need to learn how to look at the information. For example, you stated that turning off the R3 was not the right thing to do. What did you see to make that assessment? You have obviously done this more than I. The majority of the information you are providing I understand to some degree and I think with a little book work and hands on experience it will come. What I don't get is how much to crop, the RGB 16 and RGB 235, YV12 and YUY12, etc. I know that I need to dive in and try a few things.
What would a "good" histogram look like? Do you have some more examples of before and after?
I really appreciate everything that you are doing for me. I especially appreciate your patience.
Some of your questions were already answered in previous posts. But fear not. I'm collecting some color and colorspace material from earlier posts by myself and others that will answer more, and will apply to denoising as well. It's pretty busy around here, too, especially on weekend. But I'll try to post later today.
For example, you stated that turning off the R3 was not the right thing to do.
Digital R3 is not related to color balance or levels. It's a denoiser. I wrote that it works well for some tapes but is sometimes too aggressive for other tapes. It's a temporal filter that can sometimes cause motion smearing, depending on the tape.
It appears you need some catch up with basic video and graphics elements. At this point you shouldn't be in a great hurry to start modifying VHS captures wholesale. You would probably learn more quickly if you find a good encoding app and encode some AVI captures to MPEG or h.264. Or...you can load the short MPEG or mp4/h264 videos posted in these threads onto a USB flash drive or even as "data" on a DVD disc. You could play mpg on a DVD player from a data disc or from a USB input, or play mpg and mp4 off USB on a BluRay player or smart TV. You need to see what your captures actually look like on something other than your uncalibrated PC monitor.
Quote:
Originally Posted by Motleymongoose
What would a "good" histogram look like?
"good" histrogram -vs- "bad" histogram? There is no such animal. There are only histograms, period. They contain information that explains why a color balance might be incorrect and how it can be corrected -- or in the case of the sample captures shown in this forum they help to explain why some images look underexposed/dark and why others look grayed out or foggy, and what could be done to make corrections. In the case of the samples here, color correction was only a minor issue; the main problem is brightness levels.
If you have a bright daylight scene with strong sun, dark shadows, some grass, people, flowers...you'd expect a histogram pretty well populated from side to side with dark content down to black or near-black for the shadows and dark tree trunks, lots of midrange data for skin tones, grass and leaves, and brights and strong highlights for flowers and bright sky. For a night scene you'd expect most of the pixels to fall in the left or darker end of the histogram and into black, and only a few bright pixels for lighted areas or things such as street lamps or room lights. You wouldn't expect most of the histogram of a night scene to be heavily populated at the right, bright side or crowded into the middle unless the dark scene was overexposed.
Quote:
Originally Posted by Motleymongoose
What I don't get is . . . the RGB 16 and RGB 235, YV12 and YUY12, etc.
Apparently you skipped post #17, which answered some of your questions. If you don't understand the difference between luminance and color, or don't see that low RGB values are dark and high RGB values are bright, I guess you have some work ahead. Basically:
colorspace:
-----------
A digital colorspace is a way of specifying and storing luminance and chroma as numbers. For use by us humans, the data must include luminance (let's call it luma or brightness) and Red, Green, and Blue chroma. The data is stored in pixels arranged in vertical columns and horizontal lines; a 720x480 video frame is 720 columns wide and 480 horizontal lines high. These measurements are even numbers, evenly divisible by 2, 4, 8, and 16 -- or as math majors like to say, mod-2, mod-4, mod-8, mod-16. The numbers 2 and 4 will of interest here, since the data in many colorspaces are arranged in pixel groupings of 2's and/or 4's.
The colorspace that you are most familiar with (or should be, by now) is RGB. Display hardware displays luma and color as values of RGB (Red, Green, Blue) from 0 to 255 for each of the three colors. All info in RGB for all three colors is stored together in each RGB pixel. A value of zero (0) for any particular color means that there is no output of that particular Red Green or Blue color. Every color you see on display devices and transmitted in the natural world is a blend of those three colors.
RGB is used in display devices because human eyes have sensors only for Red, Green, and Blue light waves. Light waves are analog phenomena. Eyeballs have no way of seeing digital streams of 0's and 1's, and they have no YUV sensor or translation software. Eyes are RGB. Only.
The colorspace systems you'll encounter also include YUV -- Officially called YCbCr (digital) or YPbPr (analog), both of which amount to the same thing. YUV stores luma (Y) separately from a Blue-yellow channel (U, or "Blue") and a Red-Green channel (V, or "Red"). These arose from broadcasting, where transmission bandwidth for full scale RGB is really expensive and incompatible with b&w systems. Two common YUV colorspaces you'll use are YUY2 and YV12.
In YUY2, for every 2 lines of four Y luma pixels there are two U pixels on the even lines and two V pixels on the odd lines. YUY2 is used for VHS capture because YUY2 is closer to the way VHS tape stores video as YPbPr. YUY2 and other 4-to-2 colorspaces are often referred to as 4:2:2.
IN YV12, for every four Y pixels on even lines there are two U and V pixels, and for every four Y pixels on odd lines there are no U pixels or V pixels. This economy of storage follows on the fact that humans base perceptions of image "sharpness" mostly on luma and very little on chroma. YV12 is the standard colorspace for DVD and BluRay encoding. This pixel arrangement of four Y pixels to 2 chroma pixels in one group and no chroma pixels in the other group is often referred to as 4:2:0.
Because RGB stores 4 brightness and chroma values in each of every 4 pixels, such colorspaces are referred to as 4:4:4.
This cutting down on chroma pixels is called subsampling or downscaling. Restoring U and V channels into more dense chroma storage and display systems like RGB by various means of chroma interpolation is called upsampling or upscaling. Some video codecs store Y and UV pixels in different arrangements. Fortunately Avisynth is one utility that can accept many different storage systems and convert them properly to others. Most editors just screw it up, and the damage is permanent.
Because colorspace conversion involves numeric interpolation, digital rounding can lead to color corruption when you recklessly or needlessly jockey back and forth among colorspaces. It's cleaner to use only those conversions that you really need. You also have to be aware that progressive and interlaced video should be specified when making those conversions. The question newcomers always ask is Why did engineers arrange stuff like this? There are many answers to that, but the point to observe is that the arrangements are what they are and won't change anytime soon, so you have to work with them.
Avisynth works with several colorspaces, including YV12, YUY2, and RGB. Some filters and functions like Crop() or the resizers work in all three, many work in YUY2 and YV12, a few are YUY2 only, but most of the heavy hitters work only in YV12. Some, like RGBAdjust, obviously require RGB, while others like ColorYUV obviously are designed for YUV. Most of the time Avisynth will stop and tell you that you're trying to use something in the wrong colorspace. Documentation for the filter or function tells you what will work and what won't.
VirtualDub uses RGB almost exclusively, with a scant few exceptions. They also work with interlaced and non-interlaced; some will detect interlacing and act accordingly, a few will ask you to specify interlaced or not, and a very few will insist on non-interlaced. Most VDUb color filters work with both. The safest way to get a video into RGB properly is to use Avisynth Conversion functions even if you use no other Avisynth filters in the script. The safest and best way to get a video deinterlaced when required is with Avisynth. Most 3rd party editors, including some very prominent names, don't do any of these functions well. Other famous apps whose functions and color controls require RGB (most of the time) are Premiere Pro, AfterEffects Pro, and Vegas Pro, plus many of the less expensive non-pro versions of these apps like Premiere Elements and SONY Movie Studio. The expensive guys can also work with YUV.
Quote:
Originally Posted by Motleymongoose
What I don't get is how much to crop
Cropping is often for removing border areas you don't want, like bottom-border head switching noise, followed by replacing cropped border pixels with black ones. Exactly how to crop depends on the colorspace and other factors. Because of the way YUV's luma and chroma are arranged, cropping odd numbers of columns or lines will cause problems. Restrictions that apply to the Crop() function are shown at the bottom of the wiki page at http://avisynth.nl/index.php/Crop, and in Avisynth's installed online help.
Primary Colors and color blending:
---------------------------------------------
Below is a 22-ramp grayscale with colors from pure black thru shades of gray to pure white. Black is on the left and white is on the right. To relate that to dark and bright values, black is dark, mid-gray is about the middle of the spectrum, and bright white is obviously bright. RGB brightness values are marked on each vertical band.
You should be able to copy the above PNG onto your desktop at its original size, 640x180. On your monitor you should be able to differentiate clearly between all of the grayscale patches shown above, from darkest to brightest. All of the panels should appear to be neutral shades, free from easily discernible color tints or discoloration. The last 2 or 3 darkest panels and the brightest 2 or 3 bright panels should not appear to merge.
You'll notice that the panels above were described as 22 "colors". As stated earlier, Black, grays, and whites are indeed colors. Each has its own RGB values. This includes the colors "Black", "RGB 121 neutral Gray", and RGB 255 "White" above, which respectively are made from blends of primary Red, Green, and Blue, as shown below:
* Zero Black = Red 000 + Green 000 + Blue000 (i.e, pure Black has zero value in every color)
* RGB 121 Gray = Red 121 + Green 121 + Blue 121
* Bright White = Red 255 + Green 255 + Blue255.
Notice that shades of gray from black to white each contain a mix of equal parts of 3 primary colors. Will you ever find this useful? A color cast in an image will always be evident in colors that are supposed to be neutral, such as black, gray, or white. If blacks look bluish, or grays look bluish, or whites look bluish, you can be certain that there's too much Blue in the image. To correct that, many newcomers make the mistake of adding more Red. That's not correct. The way to balance too much Blue is to either reduce Blue, or to add more of its opposite colors, which would be Yellow (Red + Green).
The following users thank sanlyn for this useful post:
bar72 (08-22-2023),
bilisi (12-30-2020)
Digital R3 is not related to color balance or levels. It's a denoiser. I wrote that it works well for some tapes but is sometimes too aggressive for other tapes. It's a temporal filter that can sometimes cause motion smearing, depending on the tape.
That's correct, R3 does sharpen. Maybe the 7800 and/or 7500 is a bit different from my old 7600's, where Digital R3 had to sharpen because it did some temporal denoising and "enhancement" that softened everything and smeared motion. Could be a different effect on some machines. I recall my old JVC 9911, where R3 sharpened and also blew out highlights, which my 7600 didn't do. People who've been through several JVC models might tell us more. The 7600 and 9911 were the only ones I used.
I didn't see very much difference between the early R3 capture and the cap with R3 off. Frankly I've lately stayed away from players with image "enhancement", even Panasonic AG-1980's. Maybe those features worked better when the units were brand new, but at this point I'd rather try to live with some noise and post-processing instead of increasing levels of destruction from aging circuits. But that's just me, I guess.
I guess there are different opinions on wether or not R3 is on or off.
One other thing I would like to add to the discussion. I am using a AVT-8710 in between the VCR and the computer. I noticed that it has several adjustments. I am assuming that I want to leave it in the default setting mode. Mostly because of what you have been emphasizing, it would a one size fits all type of adjustment and destroy the good data as well as correct the bad.
Also, how much experience have you had with hi8? I may need to start another thread but any quick tips on how to archive that video. I have done it in the past through FireWire. The system that I have now does not have FireWire. Do I care?
I am going to take your advice and dig into the great information and see if I can learn something.
This is all good information in this thread, but...
I want to make sure we're not overloading the OP. Remember that just because you can restore, doesn't mean you always need to do so.
For example, I have zero interest in restoring my old little league videos. It's a straight dump to 15mbps MPEG with an ATI card, and I call it good. It looks fine. The video levels were bad, yes, but it's just as much the fault of the outdoor lighting as it is the camera (and the VHS format). Maybe if I were an all-star MLB player, yeah, it'd have some value. But I'm not, and so it doesn't.
At minimum, I hate chroma noise, so I always attack that. But the S-VHS VCRs, TBCs, and DVD recorders do a pretty good job of eliminating that.
Sometimes chroma offset is so bad that it's not enjoyable to watch, so I have to fix that.
It really comes down to "how annoying is this?"
- If not really annoying (color/levels, grain, etc), I leave it alone.
- If the audio hisses, video is unsteady, etc, then I need to fix it to enjoy it.
I guess there are different opinions on wether or not R3 is on or off.
Sometimes it gives better results, sometimes not. Sometimes it could make no difference. Every tape is a different animal. You don't have to capture hours of tape to make a test, just a few minutes will do. If there's a difference, you'll see it playing back the results. My experience is that most tapes don't benefit from sharpening during capture and can sometimes look worse. There are better enhancement filters in Avisynth and VirtualDub.
[EDIT] I had one really crummy tape that did look better with R3 turned on. But it wasn't a perfect fix. R3 enhancement also made motion smear more obvious. So it depends on which evil you want to live with. Knowing what I know now about Avisynth, it would have been better to sharpen later.
Quote:
Originally Posted by Motleymongoose
I am using a AVT-8710 in between the VCR and the computer. I noticed that it has several adjustments. I am assuming that I want to leave it in the default setting mode. Mostly because of what you have been emphasizing, it would a one size fits all type of adjustment and destroy the good data as well as correct the bad.
There is no one size fits all. I wish there was.
I've used my AVT-8710 for years. It does have some basic proc amp controls and I tried them in the past. But they're a hassle to work with, klutzy to set up. And since you don't seem to have a color problem (the problem with your caps is constricted levels) you'll have to monitor the results in a histogram anyway, especially with an uncalibrated monitor. VirtualDub capture has a histogram that works in Preview mode (but not during capture, of course). I posted a picture of a sample capture histogram in Post #11 (http://www.digitalfaq.com/forum/vide...html#post42215). VDub Capture has a "levels..." filter category that's supposed to hook into your capture card's drivers. Far and away much easier to set up than the AVT. Be careful that you don't mistake the "Levels..." menu for the "Filter chain" menu -- the latter is a no-no, as those refer to VDub RGB filters or other add-ons. Don't use those during capture.
Quote:
Originally Posted by Motleymongoose
Also, how much experience have you had with hi8? I may need to start another thread but any quick tips on how to archive that video. I have done it in the past through FireWire. The system that I have now does not have FireWire. Do I care?
You do. Hi8 is analog. Firewire is for lossy DV. You'll lose 50% of your color and find yourself fighting some noisy interlace and DV compression artifacts. But if you don't have a hi8 camcorder, DV might be your only choice.
Quote:
Originally Posted by Motleymongoose
I am going to take your advice and dig into the great information and see if I can learn something.
Well, my advice is really adapted from forum members who've done what you want to do. Take your time. Getting proper captures is the first step, you seem to be doing that okay except for learning to get more natural luminance levels that can save lots of work work later.
I agree with lordsmurf's previous post -- you can't fix everything. Even if you could, which you can't, it would take forever. The samples I posted of my sis's home videos could still use more work, but...I think the point of diminishing returns was reached by solving the worst of the problems. After all, this stuff is VHS, one of the most imperfect, contrary, and glitchy things ever devised.
Not to get too sidetracked, somewhat the same procedure for Hi8 as for VHS with the exception of going through a composite cable instead of FireWire from my camcorder?
I never used a camcorder for capturing, but those who are familiar with using one can advise. There have been many posts involving Hi8 over the years in this forum, some in recent weeks.
s-video cable, if available, is superior to composite -- yes, you can see the difference.
Not to get too sidetracked, somewhat the same procedure for Hi8 as for VHS with the exception of going through a composite cable instead of FireWire from my camcorder?
Video8, Hi8, S-VHS, and VHS have the same basic process. For example, they all need external TBC, as they're all subject to frame drops.
The main difference is chroma noise and chroma offset. The 8mm-based tapes don't really have that issue. Hi8 is better than Video8 as well, as the latter can sometimes have it.
The Hi8/Video8 tape is physically better in many cases because it wasn't compatible with a cheap VCR, thus the tapes were rarely eaten.
The downside is that VHS can be tracked, while Hi8/Video8 cannot. So if you have a misaligned signal on a Hi8/Video8 tape, and no access to the original camera (or if it has changed yet again in misalignment), then you're screwed.
