Clipping white levels (Alan...help please :) )

4 replies [Last post]
RichardB
Offline
Joined: Aug 27 1999

I've had a sudden (unpleasent) reminder that Premiere tends to output white at a higher level than is good for the old cathode ray.

As someone who does a lot of animated B&W storyboards, I USED to know at what level to clip the white output at (I think it was around the 216 mark, but now that seems v. dark.)

Any suggestions folks? Of course I want to make my graphics bright and crisp, bit also I want to deliver tapes that aren't going to give clients any problems...

Thanks in advance,

RichardB.

harlequin
harlequin's picture
Offline
Joined: Aug 16 2000

Have a look at :
http://blas.cis.mcmaster.ca/~monger/vidcolor.html

this may be of help

------------------
Gary MacKenzie
Audio Visual Technician
(email me if you want a quick reply)

Gary MacKenzie

sepulce@hotmail.com ( an account only used for forum messages )

Thinkserver TS140 , 750ti Graphics card  & LG 27" uws led backlight , Edius 8

Humax Foxsat HD Pvr / Humax Fox T2 dvbt

Alan Roberts at work
Offline
Joined: May 6 1999

Hi Richard.

DV levels conform to Rec.601. Black is at 16, white at 235, so you get 219 levels in the grey scale. So, if you have to clip, 235 is not a bad level to do it at. But, you might not like the results, to explain (now, din't you expect that to happen?):

The whole idea of video not using the full range of 256 levels is to allow head and foot room. That means filters are not forced to hard clip at black or white, they can momentarily excurse beyond, up to 255 and down to 0. But why should filters be allowed to do that anyway? Well, it's because real filters have to live in the real world, where it's not possible to implement infinite transversal filters etc, there are end effects because some terms are missing (otherwise they'd have infinite delay, put a signal in then wait for the universe to end), or if they are made using genuine hardware (coils and capacitors etc) and there's not enough sections in it. And there's the problem that real filters that do a good job in the frequency domain may wreak havoc in the temporal domain and vice versa.

As ever, we have to strike compromises between filter performance (i.e. achieving the required electrical function of defining the bandwidth) and economy (i.e. being able to build the darn thing). So, we get filters that aren't ideal, and they overshoot or ring.

I recently looked at a test card that was generated according to Rec.601 using the specifed filters to limit the bandwidth (it was generated in sofwtare, so no hardware limitations) and found that although black was at 16 and white at 235 as it should be, there were some pixels at 0 and some at 255. This was on high contrast edges, black to white and vice versa. The filters, in order to restrict the luma bandwidth to 5.5MHz and the chroma to about 2.5MHz, had to allow some overshoot. Simply clipping off those overshoots would have restored frequency components outside the permitted channel bandwidth.

So, if you have to clip, try to do it with a "soft" clipper, one that allows a few pixels through. That way you'll not put nasty high frequencies into the signal. Those frequencies may upset a vtr or transmitter and can be hard to track down.

Better to arrange things so that you don't need to clip, and that means defining white and black as 235 and 16 respectively. It was partly Premiere's apparent unawareness of matters like this that drove me towards EditDV (now CineStream) which fully understands it and allows appropriate action to be taken.

Hope that's helped.

[This message has been edited by Alan Roberts at work (edited 27 March 2001).]

RichardB
Offline
Joined: Aug 27 1999

Ahh... thanks to you both, it all comes flooding back now.... last time I had these settings it was from stumbling around with trial and error, now at least there's a bit of understanding going on...

I had completely forgotten that the black needs attention too, and will follow Alan's advice to the letter, clipping at around 235, but also pushing the black up to about 16.

Of course, black and white line drawings are exactly the situation where a lot of high contrast edges are created: I'm going to pull the median point of the curve down towards the less contrasty end of the scale, and see if that helps. Otherwise...(correct me if I'm wrong here) adjusting the gamma curve may help, no?

Having said that, I like the idea of filters with an infinite overhead... just imagine the Matrox brochere: 'Our new card no longer has to render the infinite: our new RT3000 gives you an infinite wait IN REAL TIME....'

Thanks again

RB.

Alan Roberts at work
Offline
Joined: May 6 1999

Gamma probably won't help. If you're dealing with graphics, then you've got the solution to hand, slightly soften them in your graphics package before you import them to video. Remember that it takes 3 pixels or lines to make one cycle of video, you do not have full freedom to assign any pixel to any level. Create the graphics then slightly soften and all should be well.

The way to check all this is to do a trial run. In the editor, render one frame of the material you're interested in, then export it to BMP (or whatever your favourite format is) and open that in your graphics package. Use the eyedropper (like in PhotoShop/PaintShop etc) to explore the edges and see what's happened. If you're getting significant numbers of pixels outside the 16-235 range, then slightly increase the softening you do before importing the original into video. It won't take you long to find a set of standard settings that will always do the right thing for your software.

Hope that's fixed it.