HM 700 - Triplex offset CCD

[David Parks]

According to JVC, the HM 700 uses 3 1/3 inch progressive CCD's using triplex offset.

Can someone tell us how much the pixels are offset from red,green, and blue and what pixel resolution are the CCD's? 1280x720 or 960x540?

Really curious...

[Tim Dashwood]

I was told that the CCDs are still 1280x720 and the offset is used to generate the 1920x1080p scanning instead of upscaling.

My understanding is that the blue and red CCDs are offset (probably 1/2 pixel) on different axis relative to green.

[Steve Mullen]

Originally Posted by Tim Dashwood

I was told that the CCDs are still 1280x720 and the offset is used to generate the 1920x1080p scanning instead of upscaling.

My understanding is that the blue and red CCDs are offset (probably 1/2 pixel) on different axis relative to green.

Correct. Rather than offset green 1/2 pixel horizontally and vertically as does Panasonic, JVC offsets red 1/2 pixel in one direction and blue 1/2 pixel in the other direction. JVC correctly claims this is a way of obtaining additional luma resolution. (I'd love to know HOW this differs from what Pana does.) JVC also mention a "new process" for obtaining YUV from this arrangement.

The process MAY be as important as the new arrangement. The traditional green-offset technique increases horizontal resolution by about 15% -- nowhere near the "50%" often mentioned. It is best used to add super-sampling" when the CCDs already provide resolution equal to the recorded resolution. In other words, when recording 1280x720 -- super-sampling is a great way to add a bit more horizontal detail.

Shifting green 1/2 pixel in both directions increases, by a small amount, both horizontal and vertical resolution. About 15% on each axis when measured.

However, what's left out is HOW luma is obtained. The traditional way is what I call "passive" interpolation. A simple process is used to "interpolate" the luma component.

Sony, in the V1, seems to use an "active" process that captures the 3 CMOS data to three 2GP buffers after which a DSP chip massages the data and essential creates a new set of image data. Key to this process is filling the buffers with ALL pixels at 50Hz or 60Hz. This process ups the horizontal rez by about 25% to 35% -- if I remember correctly. (And, I need to say "seems to" because after making a presentation in NYC about the chips and DSP -- the camera shipped with a manual that has different pixel numbers. And, Sony has been unwilling to re-explain the capture process given the manual's numbers.)

My point is that the process that massages chip data MAY be able to increase the amount of luma data that can be obtained above 15%. (So it would be nice to have JVC explain their process.)

However, there is the additional very real issue of how much increase can be obtained from static situations verses dynamic situations. IF, the process can obtain a 50% increase from static situations, a 1920x1080 pixel image can indeed be obtained from 1280x720 element CCDs. Such a camera should measure very well.

With motion, a 50% increase on each axis isn't very likely to be obtained. Of course, for those of us who want to shoot 720p -- this is no big deal. And, for those that want a film-look it may make no difference either since at slow shutter-speeds motion blur may be the limiting factor in apparent resolution.

Nevertheless, it is one of the reasons Sony went with FullHD CMOS chips for 1920x1080 video. However, for those who have used the EX1/3, the downside of the "greater" resolution (and increased sensitivity) is rolling-shutter. :(

PS: the role of "process" in obtaining luma rez is the equivilent of HOW de-bayering works to obtain luma with single-chip cameras. And, new "pixel arrangements" are being created for single-chip cameras: the newest Kodak arrangement tosses in a clear pixel while Sony has it's own ideas.

[David Parks]

Originally Posted by Steve Mullen

With motion, a 50% increase on each axis isn't very likely to be obtained. Of course, for those of us who want to shoot 720p -- this is no big deal. And, for those that want a film-look it may make no difference either since at slow shutter-speeds motion blur may be the limiting factor in apparent resolution.

Sounds good to me. I'm betting 720p/60 at 35 Mbits will look very clean and sharp.

Thanks Dr. Mullen and Dr. Dashwood. I have no idea what half of the pixel math is on this because, well I'm not that smart. But for a simpleton like me it sounds good.

It sounds like it should be sharper than the Panasonic HMC 150, and HPX 170.

[Carlos Corral]

Well, your also talking about 900 TV Lines with the JVC700 vs 540 Lines with the HPX170.

[Steve Mullen]

Originally Posted by David Parks

Sounds good to me. I'm betting 720p/60 at 35 Mbits will look very clean and sharp.

And, IF 1080p24, 1080p25, and 1080p30 are slightly softer, that may be very acceptable to those that shoot low-temporal rate video. To learn why, see my story in Broadcast Engineering:

24p judder

[Thomas Smet]

To explain this is more simple terms. The more native pixels you have the better results you will gain from pixel shift or whatever you want to call it. Starting at 1280x720 is already giving you a pretty solid looking image and any extra level of detail gained should at least put you in the ballpark of most 1080i HDV cameras. I'm not sure if it could ever be as raw detailed as a SONY EX series camera but I would bet my life on the JVC method being much better then trying to get to 1920x1080 from 960x540 chips. The less you have to increase the better.

The problem with normal pixel shift is that if you have colors that are made up of only blue and red then you are not really shifting anything because the red and blue are in the exact same place. The JVC method seems like in theory it may work better because each channel has it's own place and any color other then a pure solid color should benefit from the shifting.