Confused: Why are we calling this a 1920x1080 sensor?

Greg Boston · December 3rd, 2006, 01:05 PM

Quote:

Originally Posted by Tony Tremble

What are the awesome capabilities of this package as you see them?

From what I have seen there is "something" happening with the Sony V1 image. Even in a wide angle and loads of depth of field there is a real sense of depth, for want of a better explanation, a 3D feel. Is this something you've seen? I don't know how or why this should be. But it's very nice.

I know what you mean about a 3D feel. That is what good HD did for me the first time I saw it in 2002 when I got an HDTV set. My own personal theory is that you get a 3D feel from having so much more resolution available. This resolution plays not only to solid objects of varying colors, but the 'specular highlights' as well. Standard definition tv transmitted via good old NTSC has such relatively poor resolution and color saturation (due to transmission limitations), that there is no mistaking that it doesn't look real.

But when you see high definition transmitted via digital over the air and not compressed to death like it gets on the satellite providers, it has so much vivid color saturation and specular highlights that it gives a visual impression of 3D. IOW, it looks more like what your eyes see in the real world which with normal vision is 3D so your brain tricks you into perceiving it as such. Again, just my thoughts about why it looks this way, not actual scientific data.

I don't think I am allowed to say exactly what I saw at Sony HQ this last week, but all who saw it were mightily impressed.

-gb-

Tony Tremble · December 3rd, 2006, 01:15 PM

Quote:

Originally Posted by Greg Boston

I don't think I am allowed to say exactly what I saw at Sony HQ this last week, but all who saw it were mightily impressed.

-gb-

Oh no! That's worse than measurebating!!! :) :)

Come on Greg give us a hint. You know you want to.

I've never known such secrecy surrounding a camera launch.

cheers

TT

Paulo Teixeira · December 3rd, 2006, 02:33 PM

I actually like what Sony did because if you had 1920X1080 imagers being around 1/4” the low light capabilities wouldn’t be as good as it is. The Canon XH-A1/G1 does have 1440X1080 imagers that will produce a wonderful image while shooting outside but when compared to the Z1u while shooting inside with little lighting, The Z1u wins because it has bigger/less pixels per imager.

Graeme Nattress · December 4th, 2006, 11:01 AM

It's quite clear from: http://bssc.sel.sony.com/Broadcastan...U/devices.html That Sony is using an interpolation process to interpolate 1920x1080 from their 960x1080 set of data points. Due to the diagonal alignment of their pixels, extra "virtual" or interpolated data points naturally lie on the corners of the pixels in such a way as to facilitate this interpolation. Now given that for interlaced imaging, vertical resolution needs to be lowered by a filter to stop twitter, it would seem that this interpolation would introduce this filtering as a natural by-product of it's operation, thus, the 1080 vertical resolution is used to help fill in the gaps in the 960 horizontal resolution, to intelligently pad it out to 1920. You don't notice the loss in vertical resolution as that loss was necessary anyway to stop twitter. Makes total sense, and should indeed allow for a very good compromise between pixel size (ie noise levels) and resolution. Considering in most operations it will be put down to 1440x1080 with HDV compression, it would seem to be a valid compromise.

Indeed, it reminds me of the very valid compromise that Sony took with the original Z1, which was also 960x1080, but used pixel shift between the R, G and B sensors to interpolate the 1920 horizontal pixels from. Again, this allowed larger pixels, which is important when you're trying to do HD on a small chip.

Now, interpolation is not necessarily a bad word. Bayer CFA imagers use interpolation to create colour from the black and white mosaic image. This is accepted. DSLRs use Bayer CFA to do their work (apart from Sigma, but they use, you guess it, interpolation to figure out the right colours as silicon is quite a poor filter of colour), and although it's an 8mp camera or 12mp camera, we know it's all interpolated. Looks very good though. Similarly in my work at RED, where we have a 12mp 60fps CMOS sensor. To get colour images from it, guess what, we use interpolation. End results look really good though, so interpolation is not a bad word.

However, you should understand interpolation and what it means. There are various types, some better than others. I guess if you could get RAW data out of the Sony Clearvid, you could write adaptive software interpolation that could do a much better job of creating the image than the on-board DSP could, although at a much slower pace.

So, when looking at interpolated images, don't say "it's not real", but look carefully and see how real it looks. For instance, on the Sony page above, they show Sony v A v B. I think we can guess who A and B are from the descriptions, but it's obvious to me that the real 1440x1080 sensors of the Canon produce the highest resolution image. All three images have hideous levels of sharpening though, making the comparison very difficult, but the Canon does look to also have the least amount of sharpness added.

Graeme

Chris Hurd · December 4th, 2006, 01:25 PM

Many thanks as always, Graeme! Much appreciated. Compression isn't a bad word either!

;-)

Graeme Nattress · December 4th, 2006, 01:35 PM

No, compression isn't a bad word. It's a realistic compromise. Indeed, I'm told, and I wasn't there, but some top movie maker was looking at REDCODE v Uncompressed at 4k and preferred the look of the REDCOE. Now that's funny :-)

But seriously, well done compression is not a problem. And as we're finding out, when compression is done on raw data, it's much less of a problem than when done on processed images as it just holds up that much better. There's a bright future for compression I'm sure!

Graeme

Douglas Spotted Eagle · December 4th, 2006, 01:37 PM

It doesn't surprise me that a filmmaker would prefer the compressed vs uncompressed, there have been a few times I've heard shooters and editors say they prefer the look of XDCAM over the HDCAM, and there is a significant difference in compression ratios of those two cams, not to mention the colorspace.
thanks for the input, Graeme

Graeme Nattress · December 4th, 2006, 01:44 PM

XDCAM v HDCAM is not comparing apples with apples though. What I was talking about was comparing the exact same footage from the same sensor, processed identically, but one version had a lossy encode / decode step in there. With XDCAM v HDCAM there are 100 other variables that effect the image, not least that XDCAM is more modern and Sony's algorithms should have advanced in the time since HDCAM.

BTW, to my eyes, the uncompressed does look better, but I stare at it all day in super magnification as I twiddle parameters and algorithms to make it better.....

Douglas Spotted Eagle · December 4th, 2006, 01:56 PM

Yes, I know that HDCAM vs XDCAM HD isn't an apples to apples comparision, my point was that I've more than once, heard from shooters preferring the compressed vs litely compressed. They're also different in 2/3 vs 1/2 lenses, 4:2:0 vs 3:1:1, MPEG vs uncompressed amongst several other factors, as you point out.
I'm a big fan of compression, simply because of datarate savings, and the fact that we don't need to have monolithic systems to process the data. Having just picked up a Ciprico MediaVault with 4 channels of optical, I'm really respecting what compression does for us. On the other hand, uncompressed is significantly faster to work with, too.
Really looking forward to seeing your end product in the RED!

Graeme Nattress · December 4th, 2006, 02:06 PM

HDCAM is hardly uncompressed though - it's an 8bit DCT codec, just like most other codecs of it's age - DV, DVCProHD etc.

Compression is wonderful though - imagine what RED would be like without it - you'd have to be able to record about a 1GB per second at full blast to get all that information off the sensor. As we drop down to 4k30p, we add some nice compression, and now you can record it to a singel hard drive. Now that's the power of compression. Just like getting HD on a DV tape.

As for the look of compression, we're now getting philosophical..... Given compression works by removing image redundancy and things which are not easily percievable, compression can act as noise reduction and image enhancement. Most codecs work on data in a transformed space, ie DCT or DWT or whatever. Similar spaces, especially wavelet space, are used in image processing algorithms for noise reduction, for instance. SO, you could get an image coming out of mild compression actually looking better and more "together" than before it went in..... I wonder if we'll get to the point where compression is used as a positive image enhancing force....

Graeme

Greg Boston · December 4th, 2006, 06:04 PM

Quote:

Originally Posted by Graeme Nattress

I wonder if we'll get to the point where compression is used as a positive image enhancing force....

Yes Graeme, it might give new meaning to the phrase, "I squeezed the crap out of it.", if it becomes perceived as a positive. ;-)

-gb-

David Ziegelheim · December 5th, 2006, 04:43 PM

While this discussion has strayed a bit, let ask a corollary question: how does the 960x1080 sensor compare to the 1920x1080 Bayer filter CMOS sensors on a Silicon Imaging camera or a Sony A1U/HC1/HC3?

Graeme Nattress · December 5th, 2006, 05:10 PM

Tricky one..... Especially as the new Sony uses a non-Bayer pattern on it's diagonal pixels.

so, it's 960x1080x3 v 1920x1080 ? Still not really answerable without measuring, but the output of the Bayer should measure about 1440x810, and the ClearVid will probably measure around 1440x810. However, I'd hazard that the larger chip would look better.... Resolution measurements != picture quality.

Graeme

David Ziegelheim · December 5th, 2006, 10:56 PM

How did you calculate 1440x810? I thought the green would be 960x1080 and the red and blue half that. Since even the V1's image processing is 4:2:2, wouldn't only half the red and blue be used? Or in this case (960x1080), the entire sensor?

Stu Holmes · December 5th, 2006, 11:19 PM

I'd just like to say that i always find Graeme's input useful and valuable and excellent reading. Thanks Graeme for your contributions to the forum.

December 3rd, 2006, 02:33 PM	#18
Paulo Teixeira Trustee Join Date: Dec 2005 Location: Massachusetts Posts: 1,773	I actually like what Sony did because if you had 1920X1080 imagers being around 1/4” the low light capabilities wouldn’t be as good as it is. The Canon XH-A1/G1 does have 1440X1080 imagers that will produce a wonderful image while shooting outside but when compared to the Z1u while shooting inside with little lighting, The Z1u wins because it has bigger/less pixels per imager. __________________ https://www.youtube.com/user/PhotoVi...esEtc/featured https://www.pond5.com/artist/paulot

December 4th, 2006, 11:01 AM	#19
Graeme Nattress RED Problem Solver Join Date: Sep 2003 Location: Ottawa, Canada Posts: 1,365	It's quite clear from: http://bssc.sel.sony.com/Broadcastan...U/devices.html That Sony is using an interpolation process to interpolate 1920x1080 from their 960x1080 set of data points. Due to the diagonal alignment of their pixels, extra "virtual" or interpolated data points naturally lie on the corners of the pixels in such a way as to facilitate this interpolation. Now given that for interlaced imaging, vertical resolution needs to be lowered by a filter to stop twitter, it would seem that this interpolation would introduce this filtering as a natural by-product of it's operation, thus, the 1080 vertical resolution is used to help fill in the gaps in the 960 horizontal resolution, to intelligently pad it out to 1920. You don't notice the loss in vertical resolution as that loss was necessary anyway to stop twitter. Makes total sense, and should indeed allow for a very good compromise between pixel size (ie noise levels) and resolution. Considering in most operations it will be put down to 1440x1080 with HDV compression, it would seem to be a valid compromise. Indeed, it reminds me of the very valid compromise that Sony took with the original Z1, which was also 960x1080, but used pixel shift between the R, G and B sensors to interpolate the 1920 horizontal pixels from. Again, this allowed larger pixels, which is important when you're trying to do HD on a small chip. Now, interpolation is not necessarily a bad word. Bayer CFA imagers use interpolation to create colour from the black and white mosaic image. This is accepted. DSLRs use Bayer CFA to do their work (apart from Sigma, but they use, you guess it, interpolation to figure out the right colours as silicon is quite a poor filter of colour), and although it's an 8mp camera or 12mp camera, we know it's all interpolated. Looks very good though. Similarly in my work at RED, where we have a 12mp 60fps CMOS sensor. To get colour images from it, guess what, we use interpolation. End results look really good though, so interpolation is not a bad word. However, you should understand interpolation and what it means. There are various types, some better than others. I guess if you could get RAW data out of the Sony Clearvid, you could write adaptive software interpolation that could do a much better job of creating the image than the on-board DSP could, although at a much slower pace. So, when looking at interpolated images, don't say "it's not real", but look carefully and see how real it looks. For instance, on the Sony page above, they show Sony v A v B. I think we can guess who A and B are from the descriptions, but it's obvious to me that the real 1440x1080 sensors of the Canon produce the highest resolution image. All three images have hideous levels of sharpening though, making the comparison very difficult, but the Canon does look to also have the least amount of sharpness added. Graeme __________________ www.nattress.com - filters for FCP

December 4th, 2006, 01:25 PM	#20
Chris Hurd Obstreperous Rex Join Date: Jan 2001 Location: San Marcos, TX Posts: 27,368 Images: 513	Many thanks as always, Graeme! Much appreciated. Compression isn't a bad word either! ;-) __________________ CH Search DV Info Net \| 20 years of DVi \| ...Tuesday is Soylent Green Day!

December 4th, 2006, 01:35 PM	#21
Graeme Nattress RED Problem Solver Join Date: Sep 2003 Location: Ottawa, Canada Posts: 1,365	No, compression isn't a bad word. It's a realistic compromise. Indeed, I'm told, and I wasn't there, but some top movie maker was looking at REDCODE v Uncompressed at 4k and preferred the look of the REDCOE. Now that's funny :-) But seriously, well done compression is not a problem. And as we're finding out, when compression is done on raw data, it's much less of a problem than when done on processed images as it just holds up that much better. There's a bright future for compression I'm sure! Graeme __________________ www.nattress.com - filters for FCP

December 4th, 2006, 01:37 PM	#22
Douglas Spotted Eagle VASST Join Date: Jan 2004 Location: Stockton, UT Posts: 5,648	It doesn't surprise me that a filmmaker would prefer the compressed vs uncompressed, there have been a few times I've heard shooters and editors say they prefer the look of XDCAM over the HDCAM, and there is a significant difference in compression ratios of those two cams, not to mention the colorspace. thanks for the input, Graeme __________________ Douglas Spotted Eagle/Spot Author, producer, composer Certified Sony Vegas Trainer http://www.vasst.com

December 4th, 2006, 01:44 PM	#23
Graeme Nattress RED Problem Solver Join Date: Sep 2003 Location: Ottawa, Canada Posts: 1,365	XDCAM v HDCAM is not comparing apples with apples though. What I was talking about was comparing the exact same footage from the same sensor, processed identically, but one version had a lossy encode / decode step in there. With XDCAM v HDCAM there are 100 other variables that effect the image, not least that XDCAM is more modern and Sony's algorithms should have advanced in the time since HDCAM. BTW, to my eyes, the uncompressed does look better, but I stare at it all day in super magnification as I twiddle parameters and algorithms to make it better..... __________________ www.nattress.com - filters for FCP

December 4th, 2006, 01:56 PM	#24
Douglas Spotted Eagle VASST Join Date: Jan 2004 Location: Stockton, UT Posts: 5,648	Yes, I know that HDCAM vs XDCAM HD isn't an apples to apples comparision, my point was that I've more than once, heard from shooters preferring the compressed vs litely compressed. They're also different in 2/3 vs 1/2 lenses, 4:2:0 vs 3:1:1, MPEG vs uncompressed amongst several other factors, as you point out. I'm a big fan of compression, simply because of datarate savings, and the fact that we don't need to have monolithic systems to process the data. Having just picked up a Ciprico MediaVault with 4 channels of optical, I'm really respecting what compression does for us. On the other hand, uncompressed is significantly faster to work with, too. Really looking forward to seeing your end product in the RED! __________________ Douglas Spotted Eagle/Spot Author, producer, composer Certified Sony Vegas Trainer http://www.vasst.com

December 4th, 2006, 02:06 PM	#25
Graeme Nattress RED Problem Solver Join Date: Sep 2003 Location: Ottawa, Canada Posts: 1,365	HDCAM is hardly uncompressed though - it's an 8bit DCT codec, just like most other codecs of it's age - DV, DVCProHD etc. Compression is wonderful though - imagine what RED would be like without it - you'd have to be able to record about a 1GB per second at full blast to get all that information off the sensor. As we drop down to 4k30p, we add some nice compression, and now you can record it to a singel hard drive. Now that's the power of compression. Just like getting HD on a DV tape. As for the look of compression, we're now getting philosophical..... Given compression works by removing image redundancy and things which are not easily percievable, compression can act as noise reduction and image enhancement. Most codecs work on data in a transformed space, ie DCT or DWT or whatever. Similar spaces, especially wavelet space, are used in image processing algorithms for noise reduction, for instance. SO, you could get an image coming out of mild compression actually looking better and more "together" than before it went in..... I wonder if we'll get to the point where compression is used as a positive image enhancing force.... Graeme __________________ www.nattress.com - filters for FCP

December 5th, 2006, 04:43 PM	#27
David Ziegelheim Major Player Join Date: Jul 2003 Location: Warren, NJ Posts: 398	While this discussion has strayed a bit, let ask a corollary question: how does the 960x1080 sensor compare to the 1920x1080 Bayer filter CMOS sensors on a Silicon Imaging camera or a Sony A1U/HC1/HC3?

December 5th, 2006, 05:10 PM	#28
Graeme Nattress RED Problem Solver Join Date: Sep 2003 Location: Ottawa, Canada Posts: 1,365	Tricky one..... Especially as the new Sony uses a non-Bayer pattern on it's diagonal pixels. so, it's 960x1080x3 v 1920x1080 ? Still not really answerable without measuring, but the output of the Bayer should measure about 1440x810, and the ClearVid will probably measure around 1440x810. However, I'd hazard that the larger chip would look better.... Resolution measurements != picture quality. Graeme __________________ www.nattress.com - filters for FCP

December 5th, 2006, 10:56 PM	#29
David Ziegelheim Major Player Join Date: Jul 2003 Location: Warren, NJ Posts: 398	How did you calculate 1440x810? I thought the green would be 960x1080 and the red and blue half that. Since even the V1's image processing is 4:2:2, wouldn't only half the red and blue be used? Or in this case (960x1080), the entire sensor?

December 5th, 2006, 11:19 PM	#30
Stu Holmes Major Player Join Date: May 2005 Location: Europe Posts: 844	I'd just like to say that i always find Graeme's input useful and valuable and excellent reading. Thanks Graeme for your contributions to the forum.