Use of adaptive unsharp mask on RED (tm) footage.

[Dan Hudgins]

I have run a JPG->BMP version of one of a frame posted by Evin Grant through the Digital Cinema color correction I am working on in my Kinema Edit list Grading command, and fixed the flesh tone to look more like Eastman Color (tm) processed at a lab that does not use exhausted bleach. NOUSM1.JPG is with out the adaptive unsharp mask, and WUSM1.JPG is with the mask. Color masking, levels, and curves were also used. The images are 1420x710, but may have been reduced in resolution by the forum software so look a little off of the full size, if you stand back from the monitor after opening the 1024 wide images you can make a better judgment about the out-of-focus guy in the back and the effect of the unsharp mask. The results would look better I am sure if I had a 4.5k raw copy of the frame to process, but the overall effect of the RED (tm) camera's color rendering seems quite adequate... See the original on Evin Grant's post to compare...



>>Note, Some tasks or programs used require Windows ME (tm) or Windows XP Home SP3 (tm) with the FreeDOS FAT32 (tm) disk mounted as a slave drive. Things may work better if all disks are formatted FAT32, which should work up to 2TB per disk.<<

[Dan Hudgins]

Here are two sample images, NOUSM2.JPG is a color correction of Haakon's image posted by "concrete", and WUSM2.JPG is the same image with the bias set low to soften the areas of low edges, most of the noise seems to be from the JPG image source, but you should be able to see some smoothing if you stand back and view the 1024 size of the images. I was aiming for a more AGFA (tm) look to the color here, I am not sure I got that but you can compare this with the original JPG posted by "concrete"...


>>Note, Some tasks or programs used require Windows ME (tm) or Windows XP Home SP3 (tm) with the FreeDOS FAT32 (tm) disk mounted as a slave drive. Things may work better if all disks are formatted FAT32, which should work up to 2TB per disk.<<

[Gavin Greenwalt]

I don't know... looks like normal unsharp mask to me.

Now if you could tell it to only sharpen edges with a contrast < 10%, that would be something I would be interested in... (runs off to figure out how to do that.)

[Gavin Greenwalt]

Actually it turned out being pretty easy and it produces really good results!

Node tree:

Input Footage
Desaturate
Split (A | B)
(Gausian Blur B)
Merge (A | B) [Difference]
Gausian Blur
Adjust levels.
Invert

Use the above as a mask and then apply your unsharp mask. Presto! It'll now only sharpen fine details not hard edges.

[Dan Hudgins]

Thanks for the example algorithm.

The conversion to JPG makes some of the more subtle effects mushed up.

I have been working with 35mm film scans, so the idea of sharpening the low detail areas was not a goal, the grain is much too coarse, but with the RED (tm) camera it might be a good idea since the noise may be low enough.

What I did above was to -reduce- the detail in the low detail areas to take some of the JPG noise down. That can look odd sometimes.

The grain in the blue exposure on film combined with the noise from the blue in the Canon XTi (tm) seems worse than these RED (tm) camera shots (hard to tell through the JPG sometimes). I am shooting 9 exposures per frame, three R,G, and B bracketed 1/2 stop through seperation filters. A monochrome digtial camera would give lower noise for the blue with maybe 9000K+ light.

I think my Sharp/Soft filter can invert the phase of the edge detection and apply positive bias, so it might be able to do a +sharp on the low detail areas and 0% change on the edges, I will have to try that on a low noise image sometime. If not I could fiddle with the code a bit. On film that would make the grain and dust worse.

I was mostly seeing in improvement in shots made with a f/1.2 28mm to remove some of the fuzz on the near focus parts. The mask has a soft boundary to counter mate with the circle or confusion. Too much unsharp masking makes film or digital images look like bad video...

The RED (tm) camera can produce good color and sharpness, I am trying to see get some different looks from the samples...

Here is a shot of the control screen for my Sharp/Soft filter, the image was taken from "concrete"'s take on Haakon's shot...

Apparently "concrete" added some grain to this image (?), so it has more of a "film look", WUSMSS3.JPG shows the control values and the 1:1 zoom view, you can only judge filters on the 1:1 so I have a movable blow-up. WUSM3.JPG shows the result, but the effect is hard to see in the JPG... The color came out a little like "Zodiac" perhaps...

Note the waveform monitor in the bottom left corner that shows single pixel noise in the zoom area's mid-line, it is there to try to get a visual on the grain in the red, green, and blue images, on film I need to not sharpen the blue or set it to soften since the grain+sensor noise is too far out of control... The two dark gray boxes near the monotor gamma check are to set the monitor brightness, and the white and near white squares are to make sure the monitor is not clipping the whites, LCD monitors seem to mave more problems with the whites sometimes, on my CRT I can see both ends of the scale, let me know if you cannot see them, the gamma should be OK around 2.4 i.e. raw CRT or tube TV...


>>Note, Some tasks or programs used require Windows ME (tm) or Windows XP Home SP3 (tm) with the FreeDOS FAT32 (tm) disk mounted as a slave drive. Things may work better if all disks are formatted FAT32, which should work up to 2TB per disk.<<

[Dan Hudgins]

Here is my attempt to make a RED (tm) image look like something a bit after a Technicolor (tm) Dye Transfer print. I was able to see a brand new print of "The Garden of Allah (1936)" several years ago before the Dye Transfer machines were not used any more. I think it was shot on the "low speed" three strip film, the color was much better than "Gone with the Wind (1939)" or "The Wizard of Oz (1939)" which I think were shot on a higher speed film and had too much contrast and grain.

My Chroma masking is film like in that you take the would expose the red mask with the green and blue negatives, and so on, but with my control screen you can adjust the amount of RGB in each color to compensate for the crosstalk to some extent to correct for the impurity. Just working with the -Luma can make colors muddy sometimes.

Anyway, it is adjustable...

The frame was processed from the 2K image Haakon posted his link to, and downsized to 1024 wide with my Quad-pixel resize, then converted to JPG in a third party program. For internal use I have a 48bpp file type but this output was from a 24 bpp BMP file, without the unsharp mask since the 2K file had good focus it did not seem to need a unsharp mask, since if I turn the unsharp mask up enough to see through the JPG conversion it can look a little too much... To best see the affect of the filters on image, it seems best to look at the uncompressed image 1:1...

You can see in the Chroma vector 'scope display that the chroma was not clipped or pushed all the way out to the edge of the chroma hexagon.

I have to say that this shot by Haakon looks very good, and it would be hard to get film to look this good, the shading of the colors is very complex and smooth (it pains me to say that since I have worked with film most of my life)... The RED (tm) team have done a fine job...

>>Note, Some tasks or programs used require Windows ME (tm) or Windows XP Home SP3 (tm) with the FreeDOS FAT32 (tm) disk mounted as a slave drive. Things may work better if all disks are formatted FAT32, which should work up to 2TB per disk.<<

[Häakon]

I have to say that this shot by Haakon looks very good, and it would be hard to get film to look this good, the shading of the colors is very complex and smooth (it pains me to say that since I have worked with film most of my life)...
Thanks for the nice words. I shot this scene in 35mm as well (I have not yet had the time to go to telecine), so it will be interesting to see how the stock looks compared to the RED. I will post images as soon as they are available!

[Dan Hudgins]

Hi Haakon,

I am working on an 35mm scanner made from Oxberry (tm) optical printer parts, if you can send me a scrap of film maybe 6 frames long I can scan it and post the image corrected to close match of the RED (tm) frame, so we can see. The scanner works at over 3K pixels, and the 105mm printing Nikkor is sharp. My street address is at my Web site along with my personal e-mail, so you can contact me directly.

When I get more along I would like to try to color correct 2K and 4K RED (tm) frames so I can test my Kinema Edit list command with real footage from a good Digital Cinema camera.

If any others would like to see a scan of their 35mm movie film negatives you can contact me as I am interested in seeing what various camera stocks look like in sample scans. Color film tends to bleed, so it is hard to get a sharp knife edge on detail, fine grain B&W stocks can get to higher resolution, such as Sound Recording film processed in Phenidone A. This would also give you an idea of what kind of scan you could get on your own cine scanner... From what I have seen so far, the color negative is 1/2 to 1/3 as sharp as the Canon XTi (tm) on shots made with fast lenses, but better negatives shot with better lenses on slow stock may look better. The lenses used in some 35mm movie projectors probably do not resolve more that 30 line pair per mm when in focus, and more like 15 line pair/mm in their more typical out of focus condition, so with an image height of 10mm the image on the screen is about 300 pixels at 90%MTF and maybe 500 pixels at 30%MTF in your average out-of-focus multi-plex screen. Because film buckles in the projector gate, and there are no longer projectionists in the both to tuch up the focus, it is probably better most of the time not to apply an unsharp mask since a unsharp masked image that is projected out of focus quite a bit may look worse than an image that was not subjected to an unsharp mask, but I will need to do some filmout tests and check in some projection situations using both digital and film origin images. Both the unsharp mask and some resize methods can bring out the Bayer artifacts in Digital origin images, and that brings up the issue about should you use a unsharp mask before or after you resize the frame images, since the effect can come out different... anyone have thoughts on leaving images soft for better resize and projection later?

Dan

>>Note, Some tasks or programs used require Windows ME (tm) or Windows XP Home SP3 (tm) with the FreeDOS FAT32 (tm) disk mounted as a slave drive. Things may work better if all disks are formatted FAT32, which should work up to 2TB per disk.<<