Meaning that my crop integers need to be divisible by 4?
+ Reply to Thread
Results 61 to 90 of 182
So far this is giving me no problems:
p1=AviSource("Cher Rosie.avi").ConvertToYV12(interlaced=true).Trim(35 96,3727).Crop(12,4,-8,0)
p2=DirectShowSource("Cher - Rosie ODonnell Interview 1996.flv").AssumeTFF().ConvertToYV12(interlaced=fa lse).Trim(2942,3051).Crop(32,8,-32,-8)
p1.WriteHistogram("Cher Rosie.hist.txt", 10)
p2.WriteHistogram("Cher - Rosie ODonnell Interview 1996.hist.txt", 10)
ColourLike("Cher Rosie.hist.txt", "Cher - Rosie ODonnell Interview 1996.hist.txt")
MergeChroma(last,McTemporalDenoise(last, settings="very high", edgeclean=true, ecrad=3, stabilize=true, maxr=2, interlaced=true))
Last edited by Cherbette; 23rd Oct 2011 at 12:14.
However, the problem seems to have a random element, changing with minor alterations to the script, so I don't think Crop is specifically to blame (I've also had a quick look at the ColourLike source code and it seems to handle pitch correctly).
It's particularly strange that a simple AviSource().ConvertToYV12() should crash.
Perhaps there is something wrong with your Avisynth setup.
Try removing all plugins from your plugins folder except the ones actually used.
One other thing is that the utube video is 25fps and mine is 29.97 is that going to cause the two videos to be out of sync at points when building the histograms for ColourLike?
Regarding your crashes, I have discovered that for YV12 input, WriteHistogram requires the input width to be a multiple of 4. Unfortunately, if this restriction is not met, it does not report the error cleanly and screws up the exception handling mechanism, which probably explains the strange errors you were getting (although it does not explain the one with AviSource().ConvertToYV12(), where WriteHistogram was not even called. ).
So you need your Crops to leave a width that is a multiple of 4.
The individual crop values can be multiples of 2 as long as the final width is OK.
As far as WriteHistogram is concerned, the height can safely be cropped to a multiple of 2, but for other reasons interlaced video should only be vertically cropped in multiples of 4.
Here's a raw/colourlike'd before/after:
Still not perfect but I think it looks much better than it did before.
Also when I recaptured "Cher Rosie.avi" because I realized I hadn't gotten any audio, I upped the saturation a bit in my Proc Amp since my source was so washed out and desaturated. I don't know if that's a big no-no or not but somehow in my mind I figured it might help with the ColourLike function...
Last edited by Cherbette; 23rd Oct 2011 at 19:56.
Your source was too blue, the utube number was too green, together they "averaged out". Sort of. It's a shame both versions are in such bad shape.
So, what are your plans for videos in your collection that don't have comparable versions on the internet?
Considering my source was in such bad shape as far as being washed faded and desaturated, I figured this would be the best way to go since there was more color present in the Youtube video...but suggestions are welcome.
The source appears to be heavily noise filtered. There are no small details left at all. Look at Cher's hair for example. You can't see even a hint of strands. Is it that way on the tape too?
The YT version is far too yellow (people are so orange they almost glow with that hue). The YT clip looks over-filtered as well, just not as much, but at least it has some color density. It looks like a bad tape dub that was oversaturated to make it look "better". Anyway, both sources are something of a mess, there's only so much one can do with either. If the YT was made part of a collection of other clips on DVD, it would have to be resized to match the other clips -- at that point it would lose much of what it might have over the OP's original.
One can always put more time into the interview clip, especially at the YUV stage early on, before any noise reduction...but I doubt one would gain much more over the improvement cherbette has so far with this clip. Wonder why both versions have so many problems.
Last edited by sanlyn; 23rd Oct 2011 at 22:04.
I'm not really certain why the Rosie clip is in such poor shape. I imagine it's a 3rd generation tape. The guy I got it from took his originals, dubbed them into compilations and then got rid of his original masters...so mine is probably at least 3rd off. I'm not sure if his VCR has some sort of noise reduction or what but it does appear it may. Some things out there I could probably find in better quality but the Cher fans always seem to be so greedy when it comes to trading/sharing etc...
I've seen tape dubbed from cheap VCR to cheap VCR using ten-cent oem cable: no detail, no color, just clay figures. People thought that was so kewl and hi-tech (they still do) . A waste of time, technology and good plastic.
But there's not even much noise in your "original" caps. All VHS is noisey, it's a very low signal to noise system. That indicates to me that you are running some noise reduction before you capture -- maybe in the VCR or the capture driver. You should eliminate that.
I just double checked the settings on my DVDR that I'm using as a passthrough for the TBC and the NR is set to "off" and my VC500 has no NR feature so I'm not sure where it'd be coming from unless there was NR generated somewhere in the dub-off before it got to me...
The only NR I applied to the "after" video was MCTemporalDenoise on the Chroma noise set to "Very High"
Last edited by Cherbette; 24th Oct 2011 at 11:00.
All I see this "colorlike" effort doing is boosting the saturation.
Try camcorder color denoise in VDUB. Your color noise is very splotchy rather than grainy. camcorder color denoise will work well.
Also, a good NLE would be better for color correction. Color curves, saturation levels and what have you.
ok thanks...I will definitely use those tips and see what I can come up with. I'll simply use the Youtube video as a reference.
Also...I posted before about how when I recaptured the "Cher Rosie.avi" segment I doubled the "Saturation" in my Proc Amp because the source VHS is so desaturated. Is that causing more harm than good or...?
Maybe it's just the combination of the multiple generation recording, downscaling, and reencoding that makes the "original" half of the side by side example look less noisy than normal.
I sew noise in the before/after (both frames), a slight series of horizontal streaks that looks like quickie flutter. Scroll the vid faster in VirtualDub, it's more obvious. I find NeatVideo at a very low setting (very nearly "off") gets rid of it, or a low-power temporal smoother in VDub or AviSynth can clean most of it. It's for sure the clip can't stand too much denoising. It's said that most of the "detail" and grain effect in VHS is mostly noise anyway, but all that detail is missing here - making it look less noisy. But there's still some noise in there. I don't think it's enough to be visible on TV, though.
Odd-looking xclip, it looks both over-filtered and over-sharpened at the same time. But I wouldn't dare soften it.
Last edited by sanlyn; 24th Oct 2011 at 12:49.
You can load the mpg version dirtectly to test-only some filters in VDub. But if you actually decide to change anything in RGB/VDub, start with your earlier AVI (either the original or the one modified with ColourLike) rather than make changes in the MPG.
NeatVideo: you really need to find a clear area in some frame on that clip (it doesn't have to be the part of the clip you used for your MPG, it can be anywhere in the video) from which to make a noise sample patch. I haven't seen the whole clip, so I have no idea where such a clear area would be. NV wants a 132x132 pixel square of more or less neutral color with no objects in the area. See the NV manual, they have samples of where to get noise patches. Try a frame that has some titles or something, there's usually a clear area somewhere. If you just can't find anything at all, some black or dark leader would have to do (not the best), or use a smaller patch (say 64x64 pixels). NV will balk at a small patch, but you can tell NV to build the sample anyway.
If you go crazy and try to use the infamous NV defaults, go to Advanced mode and turn sharpening down to less than 15% (in fact, I'd just turn it off), then turn down the Hi, Mid and Lo filters to under 25%. AviSynth's own built-in temp smoother at low power is probably enough to make those slight streakies invisible. I just used NV on that clip at super-low as a quickie convenience to see what if it would clean 'em without destroying anything (what's to destroy? There's hardly anything there). You can set the noise filters pretty far down and just set NV's temporal filter to a mere "1" before you exit the NV interface. You really don't need much denoising, I just hate looking at those "tape streaks" whiz by.
It's really difficult to use color filters without the help of a pixel color sampler (free at http://www.netreach.net/~gavin/gavsfreeware/csamp.htm ). Amazing, what this simple tool can tell you. Use it with AviSynth color settings as well, it will read AvsPMod's viewer window ('Mod" shows pixel colors anyway in the bottom status bar). IMHO working in RGB is a good way to learn what happens with different color effects, but I believe it also makes YUV easier to understand. The big difference in RGB is that RGB lets you work with one color channel without affecting the others.
This doesn't mean that learning to fix color and levels in YUV isn't important. Check the AviSynth docs on ColorYUV http://avisynth.org/mediawiki/ColorYUV and Tweak http://avisynth.org/mediawiki/Tweak. They're mighty handy tools.
Last edited by sanlyn; 24th Oct 2011 at 14:45.
I did use the original "Cher Rosie.avi" on the left of the before and after and right side is the Gradation/Neat tweaking done to the original. I just copied source frame/output frame and put them together in photoshop.
I've got Csamp but haven't quite figured it out yet. I assume the best thing to start sampling with Csamp would be skin tone but I'm having trouble correlating the RGB values that Csamp gives for a skin tone pixel + what adjustments need to be made to correct those values Csamp gave of the skin tone. So thus far I've just been eyeballing the Gradation curves which I'm sure isn't as effective and toying around with ColorTools. I'm going to go back to one of my previous threads where you guys talked about using Csamp and how it works...
I wouldn't use Neat Video on this footage. Neat Video can be an incredible denoiser but it's always one hair away from smudging all your detail away. This footage is already soft.
in AVISynth try RemoveDirt() it's a function you have to make an AVS file for. This is very effective at removing random spots, especially in chroma, without smudging.
Then use camcorder color denoise, curves, and levels in VDUB. (I recomend using the chroma denoiser before color correction)
Your last sample looks better than the colorlike sample in my opinion. Be careful not to crush your mid-tones by pushing the contrast too much.
Anyone who sets NV's dozen or more values at just a hair away from destroying video isn't using it correctly. MCTemporalDenoise and FFT3D can do some damage, too (as I quickly discovered, first time out). Complex filters and scripts have dozens of settings that many never check, with dire results.
I believe I had the settings set fairly low for NV in that comparison...set only to filter the Chroma noise
If the images you last posted were from the original, then you're getting the hang of things.
If you look over the hmtl for the gradation curves download site http://members.chello.at/nagiller/vdub/tutorial/tutorial.html , it answers most of your questions about curves filters. A copy of that page came with the filter download. All curves work pretty much alike. whether this one or the curves from Adobe.
You might have overlooked several earlier links to sites dealing with curves, levels, and histograms, and how to use them. One of those links is on curves, here: http://www.cambridgeincolour.com/tutorials/photoshop-curves.htm . By the way, that Cambridge site has some really good info, well presented and well written. I know it's a Photoshop site, but when it comes to image adjustments for photos or video the principles are the same.
ColorMill is easier and more intuitive, not as precise as curves, but still a nice tool.
Csamp has no installer. It's a tiny free-standing app that works all by itself. Save the download to a folder, unzip it, and make a copy of Csamp.exe onto your desktop. I believe I posted images of CSamp in action in one of your earlier threads, and explained how it was used.
I have two Csamp icons on my desktop. Double-click its desktop icon, and you get the simple dialog that you saw on the download page (but it's prettier than the picture!). That little pixel info panel doesn't "stay on top", it's easily hidden behind other windows. No problem; while it's running you get a little blue icon in your right-hand taskbar. Click it when you want to use it, and the info box reappears.
As you pass your mouse pointer over your Windows apps, the info box gives pixel values for Red Green Blue. Works pretty much like the pixel readers in Photoshop, Premiere, etc., but when you let go of the mouse it goes away until you start the info panel again or click on that little blue taskbar icon.
CSamp reads a 3x3 square of pixels. Put your mouse over something that's supposed to be white. You should see three RGB values that should be equal, if its really white (pure bright white is RGB 255-255-255, but 235-235-235 also looks pretty white). On objects that are supposed to be neutral gray, you should get equal RGB numbers but with lower values (like 128-128-128, or even 94-94-94). NTSC "black" has values 16-16-16. I think you get the idea that all neutral colors from black to gray to white involve RGB values of 3 equal numbers for R, G, and B.
Basic RGB colors in Windows and most color wheels (including YUV):
Black = 0-0-0
NTSC Black 16-16-16
Middle Gray 128-128-128
NTSC White 235-235-235
Bright White 255-255-255
Flesh tones: there's a very very very general rule for checking flesh tones. First, of course, if somebody's skin looks too green or too purple, and that's what your eyes and your monitor and your tv are saying, you can bet your wallet their skin is too green or purple in that image. The question is, what are the RGB imbalances that make it look that way?
All humans, whether they are Rosie, Cher, Jackie Chan, or Morgan Freeman, have Red as the most prominent skin tone. The highest RGB color in skin is Red. But skin color also has Green and Blue. The actual mix of RGB values will vary by ethnic group, etc., but you'd be surprised how little that mix actually varies. What you look at first are bthe mid-colors in skin tone: not the shadows, not the highlights, but the tones that lie between these extremes. Men generally a higher percentage of Red than women do, while Afros have -- surprise -- dark greens and a lotta blue. By and large, anyway. I haven't measured any pics with Morgan Freeman, but obviously his RGB mix won't be like Julia Roberts'.
The highest color value will always be Red, even with pale women. Green will be from 75 to 85% of the value of Red. In turn, Blue will be 75 to 85% of the value of Green. That is a very general rule, and if you see Cher standing under a stage exploding with blue-gelled floodlights, don't expect to see those same numbers. If you'd like to go into a lot of confusing and crushing detail on the subject of skin tones (more than any of us need to know, unless we're into fashion work), try this (scroll down): http://www.photokaboom.com/photography/learn/Photoshop_Elements/color_correction/skin_...ne_samples.htm
Looking for blacks, grays, and whites tells you if your image has a color imbalance problem. If your whites say something like RGB 245-215-246, your overall image balance lacks green, at least in the brights; that means the brights are probably a bit too magenta (red + blue dominance = magenta. In terms of very bright or pale colors, that's "pink" to most of us). If you visit some of the photo color correction links we suggested, you'd know that photogs and cameramen use middle-gray cards (about RGB 128-128-128) on-scene to measure ambient color temperature and other stuff, or look for grays or whites in photos to check for same. That's because, the closer your colors get to rendering grays and whites correctly, then the closer to in-balance your overall color levels should be for the "other" colors in the image.
If that's confusing, look at it another way. Let's say you have someone in the picture who is a redhead. The hair of most redheads isn't really pure RED (although many redheads are more or less red than others). It's closer to a dull reddish orange, meaning it has some green in it to keep their hair from looking like a fire extinguisher, which is REALLY red. Some have a little more green or a little less, depending on how "Red" their hair is, and some even have a pinch of blue (which, oddly enough, will make that yellow-red hair lean toward a brownish, darker-haired redhead). But let's say that red hair on this redhead in your picture looks way too orangy-yellow than a "normal" redhead might look, maybe even toward bright brightly lit pumpkin. Remember that to get yellow you mix Red + green; to get orange, you have more red than green; to get sicky-greenish-yellow you have too much green. If that redhead's hair is turning the color of an amber squad car strobe, it has too much bright red. Now, at the same time, check out some other objects that should be a light gray around RGB 150 150 150 or a white that shows up in the neighborhood of 200 200 200 or brighter. With too much bright red in the picture, those light grays or near-whites will have high values for red; the bright colors won't be evenly distributed. If your grays and whites are too red, and your redhead's hair is too red, then you will likely find that every other bright color in the picture is too red as well. You could add more green, but that might make everything too bright. So the solution would be to cut down on the bright red.
Take a look at some darks. I happen to have a video open right now with a redhead in it. She has brownish-red hair, mostly reddish and not dark. But my CSamp says that the darkest shadow areas in her red hair are turning green. If I measure the thin black belt around her waist, it doesn't "say" black; it's actually greenish black. The "black" belt is RGB 26 37 22. Since a real "RGB 26" black should ideally measure RGB 26 26 26, that tells me that the darks are a tad short of blue, but definitely too green. That surplus of dark green explains the icky greenish shadows in the redhead's hair. So while my brights were too red, my darks are too green (and the darks could use a pinch of blue, if you're really picky). So I need less dark green down around RGB 40 to 20 or so.
In another frame of that same scene, I see a guy in a white shirt. I've already handled some of the bright colors by making brights less red, but I'm now looking at the midtone shadows in the guy's white shirt. Shadows on white clothing are various shades of gray, not darker blues, not darker reds, but gray. Grays are really just "darker shades" of white. The shirt shadows vary in an RGB range from 80 at their darkest to 135 in the lighter folds. I also notice the glow of a colored stage spotlight on one shoulder, so I'll ignore that small area because it obviously won't look pure white or pure gray. The RGB values should be equal for the various shades; for example, a "perfect" RGB 80 gray would measure 80 80 80. But they all measure with values that show too much blue in the midtones. So I should increase blue in the range from about 80 to 135, which most people would consider to be "middle blue".
How do you increase bright red without increasing dark red, how do you lower dark green without darkening the bright and middle greens, and how do you add middle blue without adding more dark or bright blue? That's where filters like curves or ColorMill come in.
ColorMill is easier and more intuitive. Its dialog window has a tab for "RED-GREEN-BLUE". Under that tab are smaller buttons for Dark, Middle, Light. So you can see that ColorMill gives you sliders for each of 3 color channels in dark, middle, and bright ranges.
ColorMill is a good starting point, but you'll find that that each dark, middle, light range isn't all that exclusive. The mids will have some slight interaction with the darks, and so forth, but the cutoff points seem to be programmed fairly well. For tweaking, curves can be more precise.
The curves dialog has a grid divided horizontally into 4 sections. It is also divided vertically, but it's those 4 horizontal bands that will get most of your attention. The horizontal band along the bottom covers the range from RGB 0 (at the very bottom) to RGB 64 at the first dividing line. The next band up covers the range RGB 64 to RGB 128. The band above that one covers RGB 128 to RGB 192. The topmost band covers 192 to 255.
(Lo and behold, and not by chance, those are the same values covered by curve controls in Photoshop, Premiere, After Effects, Vegas, and a whole bunch of other expensive guys. The bonus here is that because the VDub curves work like the big boys, we can get on the internet and look at those free tutorials and demos that tell you how to use curves. Free!).
The bottom range is what most people call "darks". The two middle bands cover, as the location implies, midtones from 64 to 192. Some of those midtones vary in brightness, but most image details - and skin tones - tend to fall in the two middle bands. Middle Gray, the spot everybody's obsessed with, is RGB 128 128 128 -- smack in the very middle of the grid. Lastly, brights are in the top band. In the real world things overlap a bit, but you get the idea.
By default, every grid for colors and/or luma in the curves control has a straight diagonal line that extends from the bottom-left corner (a brightness or intensity value of 0), crossing up and leaning toward the right, up to the rightmost upper corner at value 255. On its way, the diagonal passes exactly thru the middle point at RGB 128.
If you move any point on that line to the right, you will darken those values. If you move any point on that line to the left, you brighten those values.
Now, if I haven't put every member to sleep by now, I've definitely done it to myself. Later.
Last edited by sanlyn; 25th Oct 2011 at 07:28.
Noise Reduction Amounts = High=20, Mid=30, Low=30, Y=5, Cr=25, CB=25
Temporal Filter (On your way out of the Configure menus): Radius=1, Threshold=+40
Remember, though, I'm using a noise sample patch from another video, but one that has noise very similar to the noise I refer to, and is common with old VHS. Your mileage might vary, as any of the NV default patches might be ineffective (or too effective). Try it on a short portion, such as the medium portrait shot with Cher, and note the horizontal noise in the bright background. If your NV noise patch doesn't work in that short scene, forget it and try something else. This isn't color noise, but you still need a small amount of YCbCr setting or NV won't look for any wayward color response behind those temporal whiz-by's.