De-Bayer quality at DVinfo.net

Glenn Chan · May 21st, 2007, 01:13 PM

(Splitting this off into its own thread...)

Quote:

Originally Posted by Jason Rodriguez

It's in the playback settings of the Premiere Pro program window, and under "Global Controls". I outline this in the Readme file on the website that goes with the RAW files.

In the case of AE, there is only the high-quality setting.

Jason, thanks for the reply and pointing me towards the read me (I missed it the first time around).

I was looking at the CML Bash Greenscreen image and noticed some weirdness in the image. The black and white text (on the test chart) is not perfectly black and white. As well, there is some artifacts around some of the lines on the test chart that looks like zippering. I am looking at the RAW file in AE and am wondering if I have my system setup incorrectly or something like that. But as you have mentioned, in the case of AE there is only the high-quality setting. Any thoughts on this?

Jason Rodriguez · May 21st, 2007, 04:18 PM

There is some zippering and aliasing that is going to occur as you exceed the Nyquist limit of the sensor, and with how much high-frequency detail beyond Nyquist you allow through any filters you might have in place. There are ways to get around zippering, but it typically involves stronger filtering, so you either get some zippering with light filtering and decent sharpness, or no zippering with heavy filtering and more softness. Alternatively there are other demosaicing algorithms that are senstive to edge or direction-gradient information (zippering occurs because you are demosaicing across a high-contrast edge), so they look good on test charts and text, but have problems in real-world scenes with random detail and in situations where noise tends to throw off the edge or directional algorithms and you get "maze-patterns" or other odd artifacts where the algorithms tend to find false edges or directional patterns in natural scene images. So instead of natural noise and random pattern shaping, you have other oddities to deal with. Also threshold edge-based systems can have problems in flat areas because they may pick up and edge from other wise low-contrast detail, and create odd artifacts instead of natural noise-shaping in smooth areas.

Naturally, just like CineForm has been improving the codec over time, there will be more alternatives for demosaicing in the future (all the demosaicing algorithms right now are from CineForm). There is an open DLL format for writing demosaicing plug-ins that work with the CineForm RAW decoder, so you could feasibly write your own algorithm and have it plug directly into the workflow. If you're interested in writing an algorithm of your own, contact David Newman and I'm sure he'd be happy to give you the API specification for writing demosaicing plug-ins for CineForm RAW.

Glenn Chan · May 21st, 2007, 05:05 PM

Ah, thanks for the reply.

Jason Rodriguez · May 21st, 2007, 07:40 PM

BTW, Glenn, in order to actually see the zippering, I'm assuming you were zoomed in at 2:1 or more, right?

I've looked over the image at 1:1, and any zippering artifacts that exist on the trumpets of the Chroma Du Monde are very hard to see (and could be interpreted as a number of other things besides zippering artifacts). In other words, it seems that you have to zoom in at 2:1 or greater in order to actually confirm the zippering exists. And as noted above, the reason the zippering is there is due to the fact that we're erring on a sharper rather than softer image . . . it's not completely un-filtered (that would actually look really bad), but it's also not filtered to the point where zippering would no longer exist (and hence would look softer). But it should be at a good setting so that at a 1:1 zoom level you have a very good, naturally rendered image that's not too soft but not too sharp and artifact-ridden either.

Glenn Chan · May 22nd, 2007, 01:00 AM

Yep to see the zippering you have to be zoomed in at like 200%/300%. Though at 100%, you can still see colors on black and white elements. I would think that some real world scenes would contain fine black and white detail... newspapers, printed materials, certain fabrics, logos, etc.

May 21st, 2007, 04:18 PM	#2
Jason Rodriguez Trustee Join Date: Mar 2003 Location: Virginia Beach, VA Posts: 1,095	There is some zippering and aliasing that is going to occur as you exceed the Nyquist limit of the sensor, and with how much high-frequency detail beyond Nyquist you allow through any filters you might have in place. There are ways to get around zippering, but it typically involves stronger filtering, so you either get some zippering with light filtering and decent sharpness, or no zippering with heavy filtering and more softness. Alternatively there are other demosaicing algorithms that are senstive to edge or direction-gradient information (zippering occurs because you are demosaicing across a high-contrast edge), so they look good on test charts and text, but have problems in real-world scenes with random detail and in situations where noise tends to throw off the edge or directional algorithms and you get "maze-patterns" or other odd artifacts where the algorithms tend to find false edges or directional patterns in natural scene images. So instead of natural noise and random pattern shaping, you have other oddities to deal with. Also threshold edge-based systems can have problems in flat areas because they may pick up and edge from other wise low-contrast detail, and create odd artifacts instead of natural noise-shaping in smooth areas. Naturally, just like CineForm has been improving the codec over time, there will be more alternatives for demosaicing in the future (all the demosaicing algorithms right now are from CineForm). There is an open DLL format for writing demosaicing plug-ins that work with the CineForm RAW decoder, so you could feasibly write your own algorithm and have it plug directly into the workflow. If you're interested in writing an algorithm of your own, contact David Newman and I'm sure he'd be happy to give you the API specification for writing demosaicing plug-ins for CineForm RAW.

May 21st, 2007, 05:05 PM	#3
Glenn Chan Inner Circle Join Date: Jun 2003 Location: Toronto, Canada Posts: 4,750	Ah, thanks for the reply.

May 21st, 2007, 07:40 PM	#4
Jason Rodriguez Trustee Join Date: Mar 2003 Location: Virginia Beach, VA Posts: 1,095	BTW, Glenn, in order to actually see the zippering, I'm assuming you were zoomed in at 2:1 or more, right? I've looked over the image at 1:1, and any zippering artifacts that exist on the trumpets of the Chroma Du Monde are very hard to see (and could be interpreted as a number of other things besides zippering artifacts). In other words, it seems that you have to zoom in at 2:1 or greater in order to actually confirm the zippering exists. And as noted above, the reason the zippering is there is due to the fact that we're erring on a sharper rather than softer image . . . it's not completely un-filtered (that would actually look really bad), but it's also not filtered to the point where zippering would no longer exist (and hence would look softer). But it should be at a good setting so that at a 1:1 zoom level you have a very good, naturally rendered image that's not too soft but not too sharp and artifact-ridden either.

May 22nd, 2007, 01:00 AM	#5
Glenn Chan Inner Circle Join Date: Jun 2003 Location: Toronto, Canada Posts: 4,750	Yep to see the zippering you have to be zoomed in at like 200%/300%. Though at 100%, you can still see colors on black and white elements. I would think that some real world scenes would contain fine black and white detail... newspapers, printed materials, certain fabrics, logos, etc.