4:4:4 12-bit Uncompressed DVX100

[Juan P. Pertierra]

Yes, the speckles will be gone in the prototype. They are simply a product of the noisy experimental setup I am using right now.

I haven't had much time to find the exact size capacitor that will get rid of them completely, though they have been minimized...before there was a TON of them.

I am currently working on the DV/RAW latitude comparison....

Juan

[Justin Burris]

Juan,

The DVX/35mm comparison looks really good.

The only suggestion I would make for future comparisons like these would be to be careful about the angles. Using such a hard light, the angle of reflectance is different (on the box and bottle) from the film image to the digital one. This throws off the comparison quite a bit, because the DVX placement gets a harder reflection, and therefore blows out sooner than it would from the position where the SLR is.

A more accurate comparison could be achieved by taking the picture with the film camera, then replacing the camera with the DVX, to take a picture from exactly the same spot. Then, the lighting would be identical. Maybe I am being picky here, but I have a feeling that if the lighting was more similar here, we would be even more blown away by the similarity between the two images.

[Juan P. Pertierra]

Justin,

I know what you mean! :) I really tried to make the images as similar as possible but the two problems were:

1.Differences between the lenses and formats, the SLR had a wider field of view and no zoom, so i had to get much closer

2.The current experimental setup really doesn't let me move the DVX without making some large changes...

In a few weeks the prototype should be ready, which will allow me to make a comparison from the exact same position...

Juan

[Justin Burris]

Gotcha. Makes perfect sense.

[Juan P. Pertierra]

Latitude Comparison between 4:4:4 10-bit Uncompressed output and standard DV Output
~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Ok, this took a while so i hope it proves helpful.

An important note: this set of images is not intended to demonstrate any aspect of resolution or color, just to give an idea of the difference in dynamic range.

I compressed the file HEAVILY to make it a fast download, so the colors and resolution where very affected. The original file is 57MB, so if you would like a nicer version let me know.

http://expert.cc.purdue.edu/~pertierr/LatitudeTest.jpg

[Laurence Maher]

Wow Juan!!!!!

Incredible!!!!!

Hey, I don't know if you're hip to Scott Billup's site, but it's www.pixelmonger.com

He's a very knowledgable guy, has built his own mods, etc. You might email him there with questions about how to do tests, etc. And give hiim details about what you're doing. He's very helpful and loves pulling along the fellow man.

You probably already talk to him, but in case you don't he might have some really useful insight.

[Thomas Smet]

Not sure if this will help or not but I found a codec for free that supports 16 bit per channel as well as an alpha channel making a 64 bit video codec. It works on mac and pc with just quicktime 5. Best of all it is free. They even have a lossless codec that can get 6:1 compression with no loss but that codec is $99.00. I know it isn't 12 bit per channel but it might be an easier way for people to manage files opposed to a series of stills. Besides right now the tiff files will need to be 16 bit anyways.


http://www.digitalanarchy.com/micro/micro_none16.html

[Thomas Smet]

just adding to my last post.


the $99.00 codec will also support an audio track which is something you cannot get with tiff files. I am starting to like this Microcosm codec. The only problem is that we may not be able to write to it since it seems to be a little slow. Juan maybe you can download the free codec or the demo and see if it would work for capture. Digital Anarchy states it wasn't meant as an I/O codec but then again we arent exactly doing I/O in the normal way. Even if it doesn't work for capture the Microcosm codec sure would be nice for storing footage with 3:1 to 10:1 lossless compression. I am sure you or one of us could always write a small utility to convert tiffs to this codec once capturing is done. Even 3:1 compression would save us a lot of drive space for our projects.

[Luis Caffesse]

I just downloaded and checked out the lattitude test.
If it isn't too much trouble, I think it would be of more help to upload 2 frames that we can look at that are each properly exposed. 1 from the DV output, and 1 4:4:4 uncompressed.

From what I saw in the sample, the 4:4:4 does not seem to have any more lattitude than the DV frames. The DV seems to simply be brighter overall. I mentioned in an earlier post that I thought this was probably due to the DSP, or the DV encoding, adding a certain amount of nominal gain to the image. That is pretty common in cameras. Our '0db' settings generally don't mean there is no gain.

The 4:4:4 frames are definitely darker, but they don't seem to have any greater contrast range than the DV frames. In fact, if anything, they seem to have a lower contrast ratio. The shadows don't seem as dark, and the highlights are not as bright.

In fact, while the DV frame at about 9.6 or maybe 11 seemed to be a properly balanced and well exposed frame, none of the 4:4:4 frames seems to be properly exposed to me. Either the entire image seems too dark, or too bright and milky.

It seems that although you are dealing with more increments in the luminance due to the 12bit image (4096 vs. 256), the extremes of that range may be the same.

The ideal way to check this would be to shoot a clip chart in both DV mode, and 4:4:4 mode, each properly exposed. That way we could easily see the range in grayscale that the chips can handle.


-Luis

[Juan P. Pertierra]

Luis,

I have to say I completely disagree.

First of all, whether the frames look dark or milky is irrelevant. They are completely not color corrected. I can color correct almost all the frames in the left side to look good, while most of the DV frames have entire sections of white.

How do you explain the fact that there are DV frames which have huge areas of clipped white while the raw image still captures tons detail in that area? Remember, we are not evaluating the raw image subjectively, but rather how much information is captured. No image straight out of a CCD is going to look 'good'. Have you seen the un-corrected output from the Viper?

Finally, just drive this point home. The left image IS exactly what the camera starts with, before it processes it and outputs the image on the right to DV tape.

How exactly do you conclude there is no more latitude if the camera starts with an image with diverse information all over it, and spits out an image that is 50% white?

Maybe the complete jpg is just too compressed...the original file is 56MB in photoshop, maybe I can upload that and you can get a better look.

Juan

[Luis Caffesse]

Damn it Juan, you post too fast.
I was coming back to edit my original post.

I think the jpg is just too compressed to do any real analysis,
I started trying to compare each frame's histogram, but realized it wasn't going to tell me much. Ideally we need 2 seperate frames we can compare, each properly exposed.

Don't upload the full file for me, as I'm on a dial up and can't download something that big. I appreciate it though.


"First of all, whether the frames look dark or milky is irrelevant."

I completely agree. I was only pointing that out to talk about the range of luminance values in an image. It is not irrelevant to point out that an image is too dark if at the same time someone is pointing out the the highlights are not clipping. That was all I was saying.

"They are completely not color corrected. I can color correct almost all the frames in the left side to look good, while most of the DV frames have entire sections of white."

You are right about that, that was what I was coming back to edit my post for. I fiddled with the 4:4:4 frames for a minute, and saw I could get them looking just as good, if not better. Again, cant' tell much from that much compression though.


Do you have each of those files still saved seperately?
Perhaps you could upload the f9.6 DV frame and the f5.6 4:4:4 frame? Those seem to be somewhat close to eachother, as a good starting point.

"How do you explain the fact that there are DV frames which have huge areas of clipped white while the raw image still captures tons detail in that area?"

I thought it may have been due to the gain being added in the DV compression. Not only are teh highlights clipped, but the shadows are lighter as well. It seems the entire image was brightened, not just clipped at the highlights.

"Remember, we are not evaluating the raw image subjectively, but rather how much information is captured. No image straight out of a CCD is going to look 'good'. Have you seen the un-corrected output from the Viper?"

No I haven't, and perhaps you're right. I need to remember what it is I'm looking for, and perhaps I misunderstood the outcome of what I saw.

"The left image IS exactly what the camera starts with, before it processes it and outputs the image on the right to DV tape."

I understand that. I guess my only problem is that you seem to be comparing the DV and 4:4:4 images at the same fstop. The 4:4:4 frames are obviously darker overall, and I feel we should be comparing properly exposed frames.

Perhaps I just don't understand what to look for.
I do know that it would be great if you coudl upload those frames seperately and uncompressed (just 2 of them, not all of them).
Then maybe this will clear up for me.

In the meantime, great work.
Sorry if I confused the issue.

Keep it up.

-Luis

[Juan P. Pertierra]

Just to better illustrate the point, this site has a 'raw' image from the viper, and several steps taken to make it more film-like:

http://freespace.virgin.net/shaw.clan/dpviper.html

I say 'raw' because anyone who knows how the viper system works, by that point it already did some corrections to get it white-balanced. The original viper image is greenish, just like the raw output from any CCD.

CCD's capture LINEAR data. Until someone invents a logarithmic CCD, you WILL get 'milky' images out of the CCD because that's what a linear response to light results in.

REMEMBER: Like i warned up in the comparison image, this is ONLY for latitude and dynamic range.

The RAW data is LINEAR uncorrected data from the CCD's, while the DV OUTPUT has all the brightness, contrast, sharpness, color and film-look corrections of the DVX!

Luis:
Yes, i do have the original images but I am going to try and make a slightly bigger JPG...not sure why it got compressed that much.

Juan

[Juan P. Pertierra]

I just uploaded a much better JPG of the latitude test available at the same place as before:

http://expert.cc.purdue.edu/~pertierr/LatitudeTest.jpg

[Luis Caffesse]

Juan,

I checked out the viper link you posted, good example.

I also downloaded the new latitude film you uploaded.
MUCH nicer without all that compression.

I now see what you are saying.
The 4:4:4 at f16 is catching the highlight off the bottle, but can still be pushed to reveal all that shadow detail. The same shot in DV mode (I mean if you underexposed that much) would just crush all that shadow detail into oblivion.

Hope I didn't create confusion for anyone with my posts.
Disregard my earlier posts.

-Luis

[Juan P. Pertierra]

I was also thinking, if anyone wants...pick an exposure or two from the image, and I will color-correct the RAW images and post it in full size, together with the full size DV images.

Juan