Color correction...

[ziggy1971]

I'm trying to learn how to color correct video. I've read some tutorials like,
http://www.bubblevision.com/underwater-video/YouTube-Vimeo-levels-fix.htm
http://www.glennchan.info/articles/vegas/v8color/vegas-9-levels.htm

But I would still like some more assistance in learning this craft.

This is a screenshot of the original 720x480 (resized to 656x480 to achieve square pixels and maintain the whole image without cropping)



This is the modified image.



Steps:

1. Add Sony Levels plugin to Video track I'm working on and
   a.) adjust the Input start to lower the Black levels down to approx. 0 on the Waveform monitor
   b.) adjust the Input end to raise the Whites to approx. 255 on the Histogram monitor
   
2. Add Sony Levels plugin to Video Output FX and apply a "Computer RGB to Studio RGB" preset (to apply the effect to the final output) to change the color range from 0-255 to 16-235.

I'm wondering if someone could take a look at this and make some recommendations (I learn more from watching what people do than reading about it, sorry)

The final output video is either .mp4 or .mkv (both x264 codec, depending on whether the video will be streamed, have chapters/subtitles, etc.)

[budwzr]

The scopes just show the distribution and ranges. To actually correct, or add artistic flair, you have to get into one of the "Levels", White Balance, etc. windows.

You have to decide on a "look", then play around until you're able to create it.

Unless you're talking about broadcast standards, then that's a whole separate field.

[LMotlow]

There is plenty of software around. Vegas is one tool, there are many others. Get the principles down first.
Review: 5 Books on Video & Cinema Color Grading
Hurkman: Color Correction Handbook

[ziggy1971]

OK, I'll never be a professional colorist. Most of the videos I have look really washed out/faded. I just want to improve the look of the video while staying within the broadcast standards. Just wanted some tips on what to look for and how to try to correct them.

[budwzr]

A simple "Levels" adjustment is a start.

LEVELS ADJUSTED


KNOCK DOWN BLUE HILIGHTS A TAD


ORIGINAL

[newpball]

Basically there are two steps in color correction.

First step is to get it correct. Levels, gamma, color accuracy etc. This is pretty much a science.
Second step is the creative part where you want to play with the thing and present it to your liking. This is art, there is not a right and wrong answer here.

When I look at this video I get a "70s colorful plastic Bay City Rollers Warhol" feel about it, so I would probably 'bleach' the whites and colorize the thing some more, like so:



ORIGINAL


But again that's just an idea, there is no right or wrong about it.

[LMotlow]

@ziggy1971, I think you see the problem with working with screen captures. We'd prefer to see a piece of real, unprocessed video. The work we might suggest for these nearly unworkable screen captures would hardly be effective on the real thing.

This is a screenshot of the original 720x480 (resized to 656x480 to achieve square pixels and maintain the whole image without cropping)

I'm not sure what that's about. 656x480 is a 1.37:1 image ratio. Is that what you wanted? The image as shown has some chroma bleed, halos, and icky artifacts in some areas that might have been the result of resizing or other processing. If you are doing all this processing in the original compressed format, you're dirtying up your video.

The original pic, from the bottom image you posted. Looks dim and washed out, but not as bad as the top image. Is the top image what the original video really looked like? In your lower modified image gamma looks rather odd (Did you use brightness and contrast controls to alter this image?Don't do that any more. Use better tools). Blue looks oversaturated in the blouse of the gal on the right. Red also has a lot of noise in it. I'd think you need some cleanup before you start playing with color.


Like others said, you can make it look however you want (assuming you have the real video, not a screen cap). I came up with the mods below:


I don't use Vegas (I tried Vegas 9. Ended up with After Effects and a different encoder). The above might not be what the video is supposed to look like -- owner discretion advised, and we've never seen any of the original video to work with. I used AfterEffects Levels and Color Finesse for this, but Vegas has similar controls. Began by using a curves filter to set levels and color balance for blacks, grays, and whites. Pure bright white backgrounds tend to distract, so I set white point at RGB 230 (other color work will modify that a bit), and toned it down a bit later. But you can make it brighter if you like. With the Hue/Saturation control I lowered blue saturation. lowered Red contrast but raised middle Red saturation a little, and raised yellow saturation by about 8%. The rest was with an added curves filter and a Levels control to work with shadows and midtones, lowering the initial midpoint from 128 to about 115, which improved skin tones.

Maybe take a look at the books linked earlier. They'll save years of scouring thru web tutorials that5 won't teach you that much. Stay away from Brightness/contrast filters. Calibrate your monitor. Try to submit a piece of real, unfiltered, unresized video next time; screen caps are a pisser.

And you're right: you learn from others and by doing it.

[ziggy1971]

I do agree that providing the original (at least the one I have) would be much better.

I haven't provided the video for various reasons:

1. Copyrights
2. Spamming
3. Not knowing where to upload the video
4. Didn't want this post to become a legal vs. illegal argument
5. etc.

I'm on disability, but I love music and video. So to keep myself busy, I decided to start collecting music videos (over 4000 right now) and most of them need some kind of work, even if it's only a volume change so that I can play them randomly without having to keep changing the volume control on the stereo.

The source video is one of several thousand that I found on the net, all provided by the same person. Obviously this person recorded the videos from various TV stations. However, I cannot understand what he/she was thinking; nearly every video has various issues such as:

1. Video colors out of whack (as you can see)
2. Audio volume varies greatly from track to track
3. Audio panning (left/right) not balanced
4. Audio bitrates varying from 192-448 kbps (AC-3 2-channel)
5. Some videos are straight IVTC
6. Some are progressive with every 4th frame duplicated
7. Some are interlaced, but can be deinterlaced with Avisynth and then have every 5th frame dropped to get 23.976 fps
8. The list goes on. Pretty much the only thing the have in common is their resolution (720x480) and the original person used different versions of TMPGEnc to separate the videos.

I can provide this video in it's original .mpg format. Anyone know where I can upload it to without modifications?

I'm not sure what that's about. 656x480 is a 1.37:1 image ratio.

Well, it's quite simple really. Loosly speaking, to change the aspect ratio from DV 720x480 to 640x480, the generally accepted practice is to crop off 8 pixels off each side and resize to 640x480; I simply skip the cropping part to keep the entire image. So 640+8+8=656 (8 pixels on both sides). Think of it, most TV's crop off a certain amount around the egdes of the screen (that's why you rarely, if ever, you see the small pillars on either side of the video. Now, if you crop off of the original, how much more of the video are you missing due to various cropping effects of TV screens, etc. So, I simply try to keep the entire picture. Really, how much difference is there in actual size between 720-8-8=704 resized to 640 or simply keeping the 720 and resizing to 656? It's simply my own way of dealing with these particular videos, HD video is obviously different.

I know there are a lot if issues with the videos I have, but surely there are ways to improve them. Besides, it keeps me busy and I get to learn new things.

By the way, the only filter used was the Sony Levels filter, no Brightness & Contrast or other plugins were used. Although not color-blind, my eyesight isn't that great and have yet to learn to spot the artifacts, color bleeding and other issues mentioned. Guess you gotta know "what" you're looking for before person can start looking.

[LMotlow]

640x480 is 4:3 (aka 1.33:1). 656x480 is not 4:3. It's 1.367:1, which is wider than 4:3.
The image that's inside the black borders loooks like it would be 16:9 (1.7778:1), but it's not that wide. It's 1.74:1, which is more narrow than 16x9. Whatever....a piece of the original would tell us more.

Most new TVs allow you to disable overscan, which prevents cropping. Overscan is usually enabled by default.

It's permissible to make a short 10-second cut of video that you own, for educational purposes. Most MPG cutters will cut only on key frames, but that's not a problem in this case. You can use mpg2cut2 or DGIndex, both free. 10 seconds of MPG with motion should be enough and would be well within posting size limits here.


[EDIT] Ooops. Forgot. For what it's worth:
upper image: with borders cropped, image in frame is 628x360, or 1.74:1.
lower image: borders cropped, image as 1.777:1 (640x480, or 16:9).




Compare the drum in the middle and the speaker on the left. You can leave the side borders if you want, but the resulting image displays at the wrong aspect ratio.

[newpball]

Overscan is usually enabled by default.

Still? In 2015?

Gosh, what poverty!

Would not take me long to deduce who is responsible in maintaining this totally unnecessary and potentially image quality reducing act.

[budwzr]

[Attachment 30042 - Click to enlarge]

Like others said, you can make it look however you want (assuming you have the real video, not a screen cap). I came up with the mods below:

[Attachment 30043 - Click to enlarge]

Yah, nice work.

[ziggy1971]

I understand what is being said about 4:3 or 16:9 AR's. If I were doing this stuff for other people (home movies) or distributing these music videos to others I'd be more concerned about the AR's. But, since this is my own collection for my own enjoyment, I'm aware of the inaccuracies I'm imposing and willing to deal with it.

Just for the sake of arguement, when I import the original .mpg into Vegas, select the "Simulate Device Aspect Ration", and take a screenshot it displays at 655x480. TMPGEnc MPEG Editor 2.0 Version. 2.2.8.177 does this also.

Most new TVs allow you to disable overscan, which prevents cropping. Overscan is usually enabled by default.

My HDTV is over 8 years old, I have yet to find a feature that disables overscan.

For my own veiwing, I watch the videos on my TV via XMBC 13.2 Gotham. The PC I use has the optical link to my receiver so I can still play regular DVD's/Bluray's with full 5.1 surround.

Here is the complete, unaltered video.

Now that I figured out how to upload a video, maybe I could upload one with some serious issues...

[poisondeathray]

RE: aspect ratios -

It's a very long explanation and has been discussed in very lengthy threads here and at doom9 forum if you want more info. The short version is there are different AR "rules" or interpretations.

Vegas, TMPGEnc, Adobe Premiere, and many other programs use "Rec. ITU-R BT.601" rules . A 4:3 NTSC DVD has a pixel aspect ratio of 10:11 by these rules. 720/480 * 10/11 = 1.3636 AR, not 1.3333 AR . It's "slightly wider" than 4:3. Thus, a "square pixel" screenshot would yield 480 * 1.3636 or 654.5454 width, rounded to 655 since you can't have 1/2 pixel on a monitor, or 655x480

[hello_hello]

Basically there are two steps in color correction.

First step is to get it correct. Levels, gamma, color accuracy etc. This is pretty much a science.
Second step is the creative part where you want to play with the thing and present it to your liking. This is art, there is not a right and wrong answer here.

When I look at this video I get a "70s colorful plastic Bay City Rollers Warhol" feel about it, so I would probably 'bleach' the whites and colorize the thing some more, like so:

But again that's just an idea, there is no right or wrong about it.

Sorry, but I'd disagree. Your example there is obvious proof there's a wrong way to do something. It looks terrible. At first I thought you were having a lend....

Overscan is usually enabled by default.

Still? In 2015?

Gosh, what poverty!

Would not take me long to deduce who is responsible in maintaining this totally unnecessary and potentially image quality reducing act.

Don't keep us in suspense. Who's responsible?

Would it be the same people who realise most people don't want to watch their older DVDs with crud down each side and would prefer a nice clean picture? Or there's plenty of older video still being broadcast to which the same applies?
I think you'll find over-scanning usually can't be disabled when the TV tuner is used, and the ability to do so would otherwise be dependant on the input type and resolution. If I remember correctly a TVs "film mode" when watching 24fps Bluray video (assuming it has one) would automatically disable over-scanning. As no doubt would 3D mode. Sometimes over-scanning is appropriate. Sometimes it's not. Sometimes the user gets to choose. It's a logical concept.

[budwzr]

Yeah, Newp, your "Bleach Bypass" is horrendous. (No Offense)



This one by LMotlow is very good.

[hello_hello]

640x480 is 4:3 (aka 1.33:1). 656x480 is not 4:3. It's 1.367:1, which is wider than 4:3.
The image that's inside the black borders loooks like it would be 16:9 (1.7778:1), but it's not that wide. It's 1.74:1, which is more narrow than 16x9. Whatever....a piece of the original would tell us more.

Regardless of the aspect ratio of the picture, it's within a 4:3 frame, and without any cropping, resizing that 4:3 frame to 656x480 would be the correct aspect ratio.

It would have started out life as 720x480. Of the 720x480 resolution, 704x480 constitutes the 4:3 aspect ratio (hence the black borders down the sides). The aspect ratio when the whole 720x480 frame is resized would be around 1.366:1

Personally I'd crop the black and resize to 640x480 (I'd generally use a playback method where there's no over-scanning) although in this particular case I'd do what you've done and crop the black borders top and bottom, then adjust the cropping as required so I could resize to a 16:9 aspect ratio and the picture would fill the screen. Using the pic as a rough guide I'd go for cropping 8 pixels from both left and right, 60 pixels top and bottom, and resize to 640x360.

[LMotlow]

Vegas, TMPGEnc, Adobe Premiere, and many other programs use "Rec. ITU-R BT.601" rules . A 4:3 NTSC DVD has a pixel aspect ratio of 10:11 by these rules. 720/480 * 10/11 = 1.3636 AR, not 1.3333 AR . It's "slightly wider" than 4:3. Thus, a "square pixel" screenshot would yield 480 * 1.3636 or 654.5454 width, rounded to 655 since you can't have 1/2 pixel on a monitor, or 655x480

Of course. I've used TMPGenc Editor v2 for years and still have both copies. But the actual picture area used by the 4:3 image is 704x480. 704/480 * 10/11 = 1.33333.

TMPGenc editor display options offer a choice between DAR or source size. In any case, I think one can see that round objects in the 656x480 frame look slightly but visibly squished. The resulting image is not displayed on the TV from a 10:11 PAR but at 1:1. But the owner likes it that way. (Hmmm. Has newpball maybe met a soul mate here ? ?) Anyway, the O.P. seems aware of the difference and can handle it any way he wants.

@ziggy1971, thanks you for the sample. Viewed it twice so far. I know the colors are supposed to be kinky to suit the subject, but...wow this is a great lesson in sloppy color grading. Will have to set up some sort of graphs or pics to show what can be done. Getting kinda late here, but will be up early tomorrow. Thanks again.

[jagabo]

I dis something similar to LMotlow early this morning using AviSynth. I didn't saturate the colors as much though.

[poisondeathray]

Of course. I've used TMPGenc Editor v2 for years and still have both copies. But the actual picture area used by the 4:3 image is 704x480. 704/480 * 10/11 = 1.33333.

Yes, that's correct. It's actually supposed to be the inner 702 width for analog transfer, but most programs round it to 704 . But that's the math behind why PAR corrected screenshots taken by various editing programs usually yield 655 (sometimes rounded to 656 width), not 640 width screenshots. You're starting with 720 pixels width, even though it should be 702. It's up to the user to interpret the AR correctly in the program and according to the content - you can override the AR interpretation in most programs.

[ziggy1971]

At any rate, even if the result is not perfect, some basic modifications to the video can make it look much better (less washed out/faded).
I keep all my project files and original videos anyway, just in case I find/learn a better way of doing things.

#jagabo

I'd be interested to see the script you used in AviSynth. I do use it a lot with AvsPmod.

Some of the videos I have downloaded I also have in original DVD format (eg. Madonna Celebration DVD's) which I bought because I like many of her music videos, I could rip them and compare it to the video editing I've done to her videos. Did I mention I have a vast Original collection of videos also. There are, however, a lot of videos that are not available in-store and are nearly impossible to find on the net. For instance, can anyone find a decent version of "Lace - I Want a Man" or any other "Lace" videos?

[LMotlow]

At any rate, even if the result is not perfect, some basic modifications to the video can make it look much better (less washed out/faded).
I keep all my project files and original videos anyway, just in case I find/learn a better way of doing things.

Well, yes indeed, the sample you provided looks a whole world better than the screen cap. That will make a difference. Now that I see what the vid really does, I'd take another route than the one used on the posted image. You also have to ask, What was the creator's intention? Most music videos, including a lot of good stuff from Madonna, create an unreal world. Of course, if we don't like that "look" we can always do something about it. And there's always the chance that the shop that did the transfer to your copy could be a little careless. But now we all have a chance to take a closer look at that video in detail.

I dis something similar to LMotlow early this morning using AviSynth. I didn't saturate the colors as much though.

Now that I see the real video, I think I was pretty much off the mark. The original uses a whole different color motif than the screen cap. (motif, did he say? Pretty fancy talk for West Tennessee, LOL!).

[ziggy1971]

I also keep the border/overscan as an indicator of how much "altering" was done after recording from TV. The TV broadcasters are pretty unlikely to play any sort of video outside the "legal" Broadcast Colors (16-235). All I was trying to do is bring the video back to a realistic color range, that's all. I have no desire to even try to improve the original creators work, I like them the way "they" made them, not what I can turn them into.

A really great indicator (for me at least) is the TV-PG logo on some of the videos; it shows what the white and black should be. But hey, that's just me.

> Bjork - Human Behaviour -> looks like it wasn't really touched

I kinda doubt the original video editors/creators intended the video to look like this:
> Belphegor - Der Geistertreiber -> leaves a lot to be desired (I don't know what it means either, don't speak german)

I think you get a better idea of what kind of video I'm working with.
Thanx...

[jagabo]

#jagabo

I'd be interested to see the script you used in AviSynth. I do use it a lot with AvsPmod.

This is what I did after cropping the picture out of the Vegas screen cap:

ImageSource("_Color Correction_1.png") 
ConvertToYV12() # work in YV12

# for dark areas: only correct levels
darks = ColorYUV(cont_y=100)

# for bright areas: adjust levels, saturation, white balance
brights = ColorYUV(cont_y=100, cont_u=100, cont_v=100, off_u=7, off_v=-3)

# replace bright areas with white balanced video
Overlay(darks, brights, mask=darks.GreyScale().ColorYUV(cont_y=100))

MergeChroma(aWarpSharp(depth=20)) # sharpen chroma
ChromaShift(c=2) # shift chroma 2 pixels to the right

The general idea is that dark areas and bright areas need different treatments (in YUV). So one stream is created with adjustments for dark areas, another stream with different adjustments for bright areas. Then the two are blended together using a brightness mask to control which areas get which treatment. The ChromaShift(c=2) might be too much but with YV12 you can't shift by one pixel.

A really great indicator (for me at least) is the TV-PG logo on some of the videos; it shows what the white and black should be. But hey, that's just me.

I often use that type of thing as a guide too.

Now that I see the real video, I think I was pretty much off the mark.

Yes, without context one doesn't know what the video was supposed to look like. Even with context it's often hard to know!

[LMotlow]

I kinda doubt the original video editors/creators intended the video to look like this:
> Belphegor - Der Geistertreiber -> leaves a lot to be desired (I don't know what it means either, don't speak german)

Google Translate sez it means "The Ghost Driver".

Looks horrible. It's PAL to NTSC with blended and interlaced frames, and some animation that uses a different framing sequence. Fortunately you don't have to make it progressive to solve the levels problems. But you definitely should not resize video of this type. The blending will only look worse.

Interesting script from jagabo (thanks again!). You can try it on the Ghost Driver, but it's so badly color graded that setting levels in one scene won't work in other scenes. This one would take quite a while.

The other video has permanently blended frames from (apparently) a telecined original.

[LMotlow]

I think you get a better idea of what kind of video I'm working with.
Thanx...

Yep. Rough stuff. Look like re-encodes. The bitrate on both samples is way too low for the material, and makes the Bjork clip look soft. Got some terrific noise problems everywhere, and the low bitrates don't help:

Frame 359 of the original "Bjork" with darks lightened:


I didn't try to clear the noise or the weird frame blending -- that would be a whole different story. I touched on levels and color. Darks are crushed and/or highlights blown out in both samples. There's only so much you can retrieve with invalid video levels. The "Ghost" smple would need work scene-by-scene, no time for that now.

Top image: "Human Behavior", frame 2700 before/after. Bottom image: "Ghost Driver", frame 3166 before/after.



The attached mkv's are demos with about 1/2 the original scenes removed. I added a fade-out/fade-in between cut sections. I played with jagabo's script and a few other tricks, but it was more handy to use some "quickie" Avisynth plugins. Even with the noise cleared, you'd need more bitrate than I used here.