4:4:4 12-bit Uncompressed DVX100

Milosz Krzyzaniak · April 12th, 2004, 10:42 AM

None. E.g. Canon D300 mentioned by me has dynamic range not lower than film while in RAW mode, and I guess even better.

Obin Olson · April 12th, 2004, 12:22 PM

what?? are you sure about this? last time I looked film had like 500% more dynamic range then video...I am sure my canon 10D has less then film and it's a professional SLR 6mp camera.....

Juan P. Pertierra · April 12th, 2004, 12:28 PM

I think there are two different things we might be thinking about...one is the smallest difference in color(luminosity for one CCD/CMOS) that can be represented...in scott's book Film is shown as 16-bit RGB equivalent, so if you had 16-bit A/D's, technically you should have the same precision in luminosity,

however, another difference is what maximum level of brightness can be shown which i think is somewhat referred to as latitude(?), i.e. at which point film shows 'all white' versus a CCD. In this case, i think all CCD's and CMOS so far clip at a much lower brightness than film...i think this is in part due because film has a logarithmic response to light intensity while CCD's/CMOS are linear.

I have heard over and over, though, that the more precision in a digital aquisition system(i.e. wider a/d's) you can get a better result when applying a logarithmic(or 'S' curve) because you can pull more detail from the darks which would otherwise not be there on a lower precision system.

Once again, i'm not an expert on this, this is my best understanding of it so far....

Juan

Obin Olson · April 12th, 2004, 01:03 PM

that sounds right to me Juan...it is the brightness I was talking about.... I want to see some frame grabs at 30bit!

Juan P. Pertierra · April 12th, 2004, 02:13 PM

ok, i am almost 100% this is a 30-bit capture. Note that there are still 6 bits missing because of the limitations of my capture card, so this image has 83% of the total information.

Recall that this is 0.9 pixel aspect ratio, so it looks a bit stretched unless your viewer compensates for it.

There is a hefty amount of noise, which is NOT due to the standard thermal noise in the CCD's, but it is due to my test setup. it's less than before but still there.

I tried to make a large range of illumination, and all i can say is that on the DV footage I cannot see anything in the dark area. I can't see any thermal noise, but that is probably due to the lower 6 bits missing.

Also, this is my 'best guess' in color balance. This is probably completely off, i know nothing about colorimetry. Maybe someone can offer to put the three R,G,B frames together correctly?

http://expert.cc.purdue.edu/~pertierr/dvxcap1.tif

Milosz Krzyzaniak · April 12th, 2004, 05:37 PM

Juan.

For best comparison, please attach a 8-bit normal DV frame to EACH 10 or 12-bit frame you present. That would let us see the difeference. Just simply grab the DV footage and transform it to BMP with all the limitations and arfifacts it has.

PS. The 16-bit TIFF frame that you showed us is not correctly lit. Just pay attention to the histogram of it. most of the information is has in it's lower tonal parts. Please be more correct adjusting exposure of your dvx.

Juan P. Pertierra · April 12th, 2004, 06:02 PM

well, this is why i wanted someone who knows what they are doing to put together the image from the RGB frames...also, i think i mentioned it before, i do not have control over exposure yet, because the camera is open and that control is disconnected, so all i have is the 100W bulb i was using for lighting, and i made it as bright as possible without clipping.

Juan

Milosz Krzyzaniak · April 12th, 2004, 06:29 PM

Juan.

Just pay attention to the lighting. Can you properly read the historgram? Check it. So far the image is clearly underexposed. If you have just one light, try to take it closer to the subject so to get more lighting. And don't be afraid of overexposion as this could tell us much when comparing the RAW frame to DV frame.

And grab the DV frame for comparison.

The best test comparison frame would be preparing the scene that one part of it would be overexposed in terms of tipical DV footage; the second part of it would be quite good exposed and the third part of it - underexposed. Prepare it:
1. as a DV frame
2. as a RAW frame.

Don't care about balance as it is a second rate thing. We could correct it in photoshop for our own.

Juan P. Pertierra · April 12th, 2004, 09:16 PM

Ok, here's an attempt at what you ask.

These are exactly the same frame, captured through DV and captured RAW.

Now, along with the exposure scroller, the WB button is also disconnected so that is why it all looks so red, but you can see the difference anyway. The menu button is also disconnected so i couldn't take letterbox mode off.

I lit it from the side and had one section overexposed(in the DV viewfinder).

Note there's a lot of speckled noise in the RAW capture as usual, and this is 30bits of 36.

http://expert.cc.purdue.edu/~pertier...aptest1_DV.tif

http://expert.cc.purdue.edu/~pertier...ptest1_RAW.tif

Milosz Krzyzaniak · April 12th, 2004, 10:51 PM

Juan, thank you. That made it.

So it looks as if indeed the tonal dynamics of the RAW exeeds the DV dynamics and as hell it is a good news, making your work Juan not only a hobbyst play but a serious serious thing to turn a prosumer camera into a real high end thing.

I notice there is a lot more additional tonal range in the overexposed area than in underexposed area. Actually, the shadow information in RAW and in DV is simillar. I think this is easy to explain and is not a problem in further shooting using your equipment. The engeneers working on the camera had to make a decision of what part of this wide 12-bit tonal range put into the 8-bit DV output. What would be the choose? Of course the lower part as it makes the whole camera most sensitive.

I guess while shooting in 12-bit it would be best to overexpose the image by 1 or 2 f-stops to gain some more shadow details while still keeping some additional details in light areas over the DV. After all everything could be corrected in post production.

AND THAT IS GREAT.

Juan - tip: When you light the scene, the viewfinder of the camera as well as zebra and all the camera measure tools would be misleading. Actually to properly expose the 12-bit image you should overexpose (in terms of DV) your scene - otherwise we will not see any advantages of the 12 bit image over DV in terms of dynamic range. Just set the lighting with the HISTOGRAM!!! It is clearly seen on the histogram that the RAW image that you sent is still underexposed!!! MORE LIGHT!!!!

Of course, this will damage the DV footage but to hell with it.

Milosz Krzyzaniak · April 12th, 2004, 11:34 PM

OK I just mastered the RAW frame that you sent.

The noise. The thing I found out is that it is NOT random, some distingued areas are just clearly covered by it and others not. Actually I guess this has something to do with spatial frequrency. Those dots are more in areas where color changes most. So, is it really problem with capture card or something deeper? Would capture card react in such a way to high frequency changing not of data rate but information itself??? And pay attention that the noise is made of a pixels wich somewhat lacks R or G component. I never found a noise pixel that lacks B component!!! Maybe that could be a hint?

Juan P. Pertierra · April 12th, 2004, 11:37 PM

Milosz,

I have noticed the pattern too, so maybe it is still an intermittent connection? It seems like the speckles happen when a specific bit just goes to zero, and it seems to be only higher-end bits such that when they go to zero, it sets the color very dark...also the fact that it is a specific bit goes along with the fact that it only occurs in areas of similar color, or areas where the color changes through those values.

Maybe I can still fix this problem....if i could only find an exact pattern.

Juan

Milosz Krzyzaniak · April 13th, 2004, 12:14 AM

My suggestion is find out why this happens only to R and G components while the B component is always ok.

Jason Rodriguez · April 13th, 2004, 04:45 AM

Actually you don't need to overexpose the camera by two stops, that's the whole point of using a higher bit-depth image. There's a TON of shadow detail to lift up if you apply the correct curve to the image. Using curves, you can get a really nice image that keeps both the shadows and the highlights. Also if I'm not mistaken, there's probably another stop in the shadows if he's missing the lower two bits of the image. Either way, I think you should be using the zebras, as they will give you an exposure that will capture all the shadow detail without burning out the highlights-maybe you could overexpose by half-a stop over 100 zebra, but I wouldn't go much futher than that, because you don't know when the highlights in the RAW files will burn out, and when they do that's bad. And you don't have to worry about banding with the 12-bit images when you "lift" up the shadows. So I don't really see the harm in underexposing by a bit.

James Ball · April 13th, 2004, 07:33 AM

http://www.dv.com/jive3/thread.jspa?...1061&tstart=15

http://home.teleport.com/~gdi/vancecam.htm

Here’s a guy who, like many of us, thought the off the shelf offerings just wouldn’t hack it.

His approach is a bit different but he has a lot of good ideas. His approach is to treat the camcorder for what it is a cobbled together collection of technology, a systems approach.

When he finished he had something that was actually useable.

Just a bit off topic because it's not a DVX100 based system but its cool none the less.

April 13th, 2004, 07:33 AM	#315
James Ball Regular Crew Join Date: Dec 2003 Location: Sulphur, LA Posts: 61	not a DVX100 mod but cool none the less http://www.dv.com/jive3/thread.jspa?...1061&tstart=15 http://home.teleport.com/~gdi/vancecam.htm Here’s a guy who, like many of us, thought the off the shelf offerings just wouldn’t hack it. His approach is a bit different but he has a lot of good ideas. His approach is to treat the camcorder for what it is a cobbled together collection of technology, a systems approach. When he finished he had something that was actually useable. Just a bit off topic because it's not a DVX100 based system but its cool none the less.

April 12th, 2004, 10:42 AM	#301
Milosz Krzyzaniak Regular Crew Join Date: Mar 2004 Location: Germany Posts: 64	None. E.g. Canon D300 mentioned by me has dynamic range not lower than film while in RAW mode, and I guess even better.

April 12th, 2004, 12:22 PM	#302
Obin Olson Trustee Join Date: Jan 2003 Location: Wilmington NC Posts: 1,414	what?? are you sure about this? last time I looked film had like 500% more dynamic range then video...I am sure my canon 10D has less then film and it's a professional SLR 6mp camera.....

April 12th, 2004, 12:28 PM	#303
Juan P. Pertierra Major Player Join Date: Jul 2003 Posts: 479	I think there are two different things we might be thinking about...one is the smallest difference in color(luminosity for one CCD/CMOS) that can be represented...in scott's book Film is shown as 16-bit RGB equivalent, so if you had 16-bit A/D's, technically you should have the same precision in luminosity, however, another difference is what maximum level of brightness can be shown which i think is somewhat referred to as latitude(?), i.e. at which point film shows 'all white' versus a CCD. In this case, i think all CCD's and CMOS so far clip at a much lower brightness than film...i think this is in part due because film has a logarithmic response to light intensity while CCD's/CMOS are linear. I have heard over and over, though, that the more precision in a digital aquisition system(i.e. wider a/d's) you can get a better result when applying a logarithmic(or 'S' curve) because you can pull more detail from the darks which would otherwise not be there on a lower precision system. Once again, i'm not an expert on this, this is my best understanding of it so far.... Juan

April 12th, 2004, 01:03 PM	#304
Obin Olson Trustee Join Date: Jan 2003 Location: Wilmington NC Posts: 1,414	that sounds right to me Juan...it is the brightness I was talking about.... I want to see some frame grabs at 30bit!

April 12th, 2004, 02:13 PM	#305
Juan P. Pertierra Major Player Join Date: Jul 2003 Posts: 479	ok, i am almost 100% this is a 30-bit capture. Note that there are still 6 bits missing because of the limitations of my capture card, so this image has 83% of the total information. Recall that this is 0.9 pixel aspect ratio, so it looks a bit stretched unless your viewer compensates for it. There is a hefty amount of noise, which is NOT due to the standard thermal noise in the CCD's, but it is due to my test setup. it's less than before but still there. I tried to make a large range of illumination, and all i can say is that on the DV footage I cannot see anything in the dark area. I can't see any thermal noise, but that is probably due to the lower 6 bits missing. Also, this is my 'best guess' in color balance. This is probably completely off, i know nothing about colorimetry. Maybe someone can offer to put the three R,G,B frames together correctly? http://expert.cc.purdue.edu/~pertierr/dvxcap1.tif

April 12th, 2004, 05:37 PM	#306
Milosz Krzyzaniak Regular Crew Join Date: Mar 2004 Location: Germany Posts: 64	Juan. For best comparison, please attach a 8-bit normal DV frame to EACH 10 or 12-bit frame you present. That would let us see the difeference. Just simply grab the DV footage and transform it to BMP with all the limitations and arfifacts it has. PS. The 16-bit TIFF frame that you showed us is not correctly lit. Just pay attention to the histogram of it. most of the information is has in it's lower tonal parts. Please be more correct adjusting exposure of your dvx.

April 12th, 2004, 06:02 PM	#307
Juan P. Pertierra Major Player Join Date: Jul 2003 Posts: 479	well, this is why i wanted someone who knows what they are doing to put together the image from the RGB frames...also, i think i mentioned it before, i do not have control over exposure yet, because the camera is open and that control is disconnected, so all i have is the 100W bulb i was using for lighting, and i made it as bright as possible without clipping. Juan

April 12th, 2004, 06:29 PM	#308
Milosz Krzyzaniak Regular Crew Join Date: Mar 2004 Location: Germany Posts: 64	Juan. Just pay attention to the lighting. Can you properly read the historgram? Check it. So far the image is clearly underexposed. If you have just one light, try to take it closer to the subject so to get more lighting. And don't be afraid of overexposion as this could tell us much when comparing the RAW frame to DV frame. And grab the DV frame for comparison. The best test comparison frame would be preparing the scene that one part of it would be overexposed in terms of tipical DV footage; the second part of it would be quite good exposed and the third part of it - underexposed. Prepare it: 1. as a DV frame 2. as a RAW frame. Don't care about balance as it is a second rate thing. We could correct it in photoshop for our own.

April 12th, 2004, 09:16 PM	#309
Juan P. Pertierra Major Player Join Date: Jul 2003 Posts: 479	Ok, here's an attempt at what you ask. These are exactly the same frame, captured through DV and captured RAW. Now, along with the exposure scroller, the WB button is also disconnected so that is why it all looks so red, but you can see the difference anyway. The menu button is also disconnected so i couldn't take letterbox mode off. I lit it from the side and had one section overexposed(in the DV viewfinder). Note there's a lot of speckled noise in the RAW capture as usual, and this is 30bits of 36. http://expert.cc.purdue.edu/~pertier...aptest1_DV.tif http://expert.cc.purdue.edu/~pertier...ptest1_RAW.tif

April 12th, 2004, 10:51 PM	#310
Milosz Krzyzaniak Regular Crew Join Date: Mar 2004 Location: Germany Posts: 64	Juan, thank you. That made it. So it looks as if indeed the tonal dynamics of the RAW exeeds the DV dynamics and as hell it is a good news, making your work Juan not only a hobbyst play but a serious serious thing to turn a prosumer camera into a real high end thing. I notice there is a lot more additional tonal range in the overexposed area than in underexposed area. Actually, the shadow information in RAW and in DV is simillar. I think this is easy to explain and is not a problem in further shooting using your equipment. The engeneers working on the camera had to make a decision of what part of this wide 12-bit tonal range put into the 8-bit DV output. What would be the choose? Of course the lower part as it makes the whole camera most sensitive. I guess while shooting in 12-bit it would be best to overexpose the image by 1 or 2 f-stops to gain some more shadow details while still keeping some additional details in light areas over the DV. After all everything could be corrected in post production. AND THAT IS GREAT. Juan - tip: When you light the scene, the viewfinder of the camera as well as zebra and all the camera measure tools would be misleading. Actually to properly expose the 12-bit image you should overexpose (in terms of DV) your scene - otherwise we will not see any advantages of the 12 bit image over DV in terms of dynamic range. Just set the lighting with the HISTOGRAM!!! It is clearly seen on the histogram that the RAW image that you sent is still underexposed!!! MORE LIGHT!!!! Of course, this will damage the DV footage but to hell with it.

April 12th, 2004, 11:34 PM	#311
Milosz Krzyzaniak Regular Crew Join Date: Mar 2004 Location: Germany Posts: 64	OK I just mastered the RAW frame that you sent. The noise. The thing I found out is that it is NOT random, some distingued areas are just clearly covered by it and others not. Actually I guess this has something to do with spatial frequrency. Those dots are more in areas where color changes most. So, is it really problem with capture card or something deeper? Would capture card react in such a way to high frequency changing not of data rate but information itself??? And pay attention that the noise is made of a pixels wich somewhat lacks R or G component. I never found a noise pixel that lacks B component!!! Maybe that could be a hint?

April 12th, 2004, 11:37 PM	#312
Juan P. Pertierra Major Player Join Date: Jul 2003 Posts: 479	Milosz, I have noticed the pattern too, so maybe it is still an intermittent connection? It seems like the speckles happen when a specific bit just goes to zero, and it seems to be only higher-end bits such that when they go to zero, it sets the color very dark...also the fact that it is a specific bit goes along with the fact that it only occurs in areas of similar color, or areas where the color changes through those values. Maybe I can still fix this problem....if i could only find an exact pattern. Juan

April 13th, 2004, 12:14 AM	#313
Milosz Krzyzaniak Regular Crew Join Date: Mar 2004 Location: Germany Posts: 64	My suggestion is find out why this happens only to R and G components while the B component is always ok.

April 13th, 2004, 04:45 AM	#314
Jason Rodriguez Trustee Join Date: Mar 2003 Location: Virginia Beach, VA Posts: 1,095	Actually you don't need to overexpose the camera by two stops, that's the whole point of using a higher bit-depth image. There's a TON of shadow detail to lift up if you apply the correct curve to the image. Using curves, you can get a really nice image that keeps both the shadows and the highlights. Also if I'm not mistaken, there's probably another stop in the shadows if he's missing the lower two bits of the image. Either way, I think you should be using the zebras, as they will give you an exposure that will capture all the shadow detail without burning out the highlights-maybe you could overexpose by half-a stop over 100 zebra, but I wouldn't go much futher than that, because you don't know when the highlights in the RAW files will burn out, and when they do that's bad. And you don't have to worry about banding with the 12-bit images when you "lift" up the shadows. So I don't really see the harm in underexposing by a bit.