"The Spatial Offset Ruse"

[Eddie B]

In Barry Braverman's book, Video Shooter, he has a heading "The Spatial Offset Ruse" which I quote partially here.

To suppress...artifacts, engineers adopted a strategy of spacial offset wherby the green sensor is offset one-half pixel with respect to the red and blue...

The spatial offset ruse is not without its drawbacks. As a National Geograpraphic shooter, I can recall shooting many high-detail landscapes that contained mostly green tones. Give the spatial offset approach, a large portion of such images as a grassy meadow or rainforest would not be offset and thus subject to substantial artifacts and loss of image quality (Conversely, the same can also be said for images containing very little green, as such scenes would also not be offset and thus subject to a loss of resolution.)...

Shooters with a penchant for capturing high-detail landscapes may want to investigate the spatial offset characteristic of their camera. While the amount of offset is fixed at the time of chip manufacture, some mitigation of the artifacts caused by too much or too little offset may be possible by adjusting the detail level in the camera or adding a physical glass camera filter--two key strategies that should be foremost in the mind of every shooter anyway.

You can read the full section with visual aids here.



For the most part, I understand what's being said. I'm very curious about how accurate this is with regards to the DVX100 specifically and some other tips or commentary that the veteren professionals here can offer.

 

[Barry_Green]

It's a good bet that every 3-chip camera on the market is now using spatial offset. It substantially increases resolution by giving the sensors a sensitive area on the between-pixel sections of the chip.

Some cameras make more use of it than others (the HVX200/HPX170/HMC150/HPX500 series, for example). But they all use it.

About the only cameras that don't use spatial offset are single-chip Bayer pattern sensors. And they can't use spatial offset because they only have one chip. They instead have lower actual resolution as compared to a comparable 3-chip design. The Bayer system must be de-mosaic'd, resulting in a similar type of effect as spatial offset does.

If your camera is employing spatial offset to compensate for resolution that isn't there otherwise, then yes, Mr. Braverman's observations are correct, and widely detailed leafy green landscapes and meadows on an HVX aren't going to be as crisp as they would be on an EX1, no argument whatsoever. In general, spatial offset increases the overall luminance resolution of a system by a maximum of 50%, with a more likely real-world maximum of around 41%.

If your sensors have enough pixels to resolve the detail as-is, then spatial offset serves primarily to help overcome the fixed-pattern noise and striping that happens on non-offset sensors. If your sensors have lower resolution than the target frame, spatial offset helps create that additional resolution and literally "fills in the gaps" to bring the sensor block's resolution up to fill out the target frame.

Referencing a DVX100, it's a non-issue. The DVX uses spatial offset, just like everyone else does. Its chips are native frame resolution, so it's got the ability to resolve full detail in all color channels. However, it uses 4:1:1 color sampling, which reduces color accuracy in the horizontal direction. Fields of grass, etc., should look every bit as good on a DVX as on any other SD camera, but chroma keyed edges won't be as good as they would be on a 4:2:2 camera. That's not due to spatial offset, that's due to color sampling.

 

[Eddie B]

I guess where I'm most confused comes from my understanding of this section of the book. My understanding of what he wrote (and it seems to be explicit enough that I couldn't have misinterpreted) is that in every shot, you need to have greens, reds and blues.

The book uses a closeup of a dollar bill as an example and claims that if you shoot an all-green object, your image will be pixelated and the only way to resolve this issue is to make sure you get reds and blues in every shot.

I guess my concern and the point of this thread is to find out how true this is and to what degree it should be a concern. Should shooters superficially add items of blue, green and red to every single shot? In my heart, this seems like an exaggerated over-reaction but at the same time, this is exactly my understanding of what Braverman is talking about.


Personally, I have never noticed any problem from not having something from every color in the frame and most people would suggest that's enough of an answer in itself but I'm trying to become an educated and better shooter and I don't want to be apathetic when I'm presented with information like this.

 

[Eddie B]

And thank you very much for your detailed response, prior.

 

[Eddie B]

If your camera is employing spatial offset to compensate for resolution that isn't there otherwise, then yes, Mr. Braverman's observations are correct, and widely detailed leafy green landscapes and meadows on an HVX aren't going to be as crisp as they would be on an EX1, no argument whatsoever.

Reading over your reply a few more times I think I understand this better but let me clarify: as long as you keep your detail down, you should be safe in regard to this subject at hand? In other words, this only becomes an issue when you begin to crank up detail and everything becomes pseudo-sharp?

This makes a lot of sense, if that's what you're saying.

 

[Barry_Green]

is that in every shot, you need to have greens, reds and blues.

That is correct. But the world at large pretty much takes care of that for us. There are very, very, very, very few instances where something is absolutely perfectly completely green, with no blue or red component in it. And, note, we're not talking about a scene, we're talking about each individual pixel. It's not that "each scene" needs to have something red or something blue in it. It's that when an absolutely pure green (or red, or blue) object is seen by the sensor, it will be seen at that sensor's resolution and no higher. Whereas an object's color includes some degree of one or the other color in it as well, it will be sensed by all three sensors and the spatial offset magic can then come into play and give the higher resolution.

and claims that if you shoot an all-green object, your image will be pixelated and the only way to resolve this issue is to make sure you get reds and blues in every shot.

This is true. But it's largely a non-issue. Where you're going to find far more occurrences of problems with pixelization is if you shoot under forced lighting conditions of all-red or all-blue. This happens to concert/dance/performance shooters all the time, the lighting director will put out a "red wash" or a "blue wash" and the image goes all to hell, becoming blocky and soft. It's not the spatial offset that's causing the problem, it's the low color sampling of the DV codec.

I guess my concern and the point of this thread is to find out how true this is and to what degree it should be a concern.

It's true, and it's pretty much of no concern. With a DVX, it's never of concern. You have to understand that all video codecs throw away a minimum of 50% of the recorded color information anyway; with DV it's 75% that gets thrown out. The only time that this would be a compelling issue would be if you were recording 4:4:4 color and you were recording a truly isolated color, a pure blue or a pure red or a pure green.

Or, as prior said, it can become more of an issue if you're dealing with an undersampled sensor which is using the spatial offset technique to boost the resolution of the image.

Should shooters superficially add items of blue, green and red to every single shot?

Definitely not. The scene is what it is.

 

[Eddie B]

Okay, I think I understand it as much as is practically needed for me to. Thanks for your help. I do have one question, just out of general curiosity, and then we can wrap this topic up.

There are very, very, very, very few instances where something is absolutely perfectly completely green, with no blue or red component in it. And, note, we're not talking about a scene, we're talking about each individual pixel. It's not that "each scene" needs to have something red or something blue in it. It's that when an absolutely pure green (or red, or blue) object is seen by the sensor, it will be seen at that sensor's resolution and no higher. Whereas an object's color includes some degree of one or the other color in it as well, it will be sensed by all three sensors and the spatial offset magic can then come into play and give the higher resolution.

Is this why the DVX doesn't have an option for shooting black and white? I know in your book you mention that with saturation at its lowest, it is still in color and the DVX can never be used to shoot black and white with in camera settings. Is that because it NEEDS the colors to function properly?

And at this point I'm slightly changing subjects, but if that theory of mine is true should there be a concern of shooting with saturation at its lower settings? Unlike most people it seems, I much prefer shooting with my settings all set to give me the most faded, desaturated colors possible. Is that going to be at the cost of resolution? I know that too much saturation is dangerous and in addition to personal visual preference, I thought that desaturating as much as possible would also give me more safety from those notorious pixelated reds.

 

[Barry_Green]

No, not related. The spatial offset process happens as the chips are being sampled into the digital domain. The color manipulations (chroma level, etc) happen after that. So it's already sampled.

In B&W it'd look pretty darn good, actually, it should deliver as much luminance resolution as a Digital Betacam.

 

[Eddie B]

So then is my assumption about shooting with settings for the lowest saturation possible for a better picture (a factual better picture, not better in terms of personal preference for saturation) accurate?

 

[Barry_Green]

If you prefer desaturated colors, go right ahead. I like the opposite, I love a richly saturated punchy look. Neither will change the recorded resolution of the image. However, in analog NTSC video, you do want to avoid oversaturation because NTSC can't deal with it very well.

 

[Eddie B]

I guess resolution is the wrong word; I'm specifically talking about the best way to deal with reds when they are absolutely unavoidable.

 

[Barry_Green]

Red and Blue are going to be severely limited resolution on any DV camera. The color sampling in NTSC DV is 180x480. If you run into strong pure reds, they're going to be blocky and there's nothing you can do about it, whether you're using a DVX or a $15,000 DSR570. If it's DV, it's 4:1:1, and that means the reds and blues are encoded at 1/4 resolution.

You can run a "chroma blur" filter on your footage in post and that will help clean up the edges some. In Vegas set it to blur by 2.0 pixels in the horizontal and 0.0 in the vertical and it can somewhat help.

 

[DPStewart]

Wow.
That was deep.