Chroma Subsampling

Started Jan 11, 2017 | Discussions
guzzibreva1 Senior Member • Posts: 1,076
Chroma Subsampling

http://users.wfu.edu/matthews/misc/jpg_vs_gif/JpgCompTest/JpgChromaSub.html

I remember doing an experiment with the Heath Kit color television I built a long time ago.  I disconnected the delay line for the brightness signal leaving only the chroma information to be displayed on the screen.  I was amazed at how little color sharpness there was.

Tom Schum
Tom Schum Veteran Member • Posts: 8,870
Re: Chroma Subsampling
1

I used to do TV repair, and once upon a time I also built a Heathkit color TV. I never did the experiment of cutting off the luminance channel, unfortunately.

Color TV was a brilliant compromise in which the color information was added to the existing monochrome TV standard in such a way that the existing base of many millions of monochrome TV sets still worked.

The compromise was based on color perception science, which showed that for some colors the human eye had less acuity than for others. Additionally the target screen size was the large (for then) 25" diagonal 4x3 screen at a distance of 8-10 ft.

This is smaller than a 4x6 photo viewed at 12 inches distance.

I can remember getting up close to the screens of these TV sets (after careful convergence and alignment) and seeing that the edge details always went to gray.

The link you provided was very informative, but it deals with standard color filter array sensor processing. A single "pixel" in such an image is actually an interpolated result of comparing that pixel (with its color filter) to adjacent pixels (with their different color filters). The resolution at pixel level is not equal to the luminance resolution possible, partly because the color filters themselves confuse the luminance information. Both luminance and chrominance have to be derived by interpolation. Single pixel resolution is not possible with interpolation, since more than one pixel is involved in the determination of detail, whether it is chrominance or luminance. Some will dispute this of course.

In the link, single pixel details were mangled but the test was intrinsically flawed by the fact that a color filter array sensor was most likely used to generate the test image.

The only digital imaging system that is able to deliver chrominance resolution equal to luminance resolution is the Sigma Foveon, as seen in classic Foveon and in Merrill, but not necessarily in Quattro, their latest offering. Of course, some will dispute this.

However, such an image might survive the processing quite well.

-- hide signature --

Tom Schum
Every day a new image.

 Tom Schum's gear list:Tom Schum's gear list
Fujifilm X30 Sigma dp0 Quattro Fujifilm X-E1 Sigma sd Quattro Fujifilm X-T3 +10 more
Scottelly
Scottelly Forum Pro • Posts: 14,217
Re: Chroma Subsampling

Interesting! Thank you for that link. I will remember to turn off chroma subsampling, because I am not that concerned with file size these days.

-- hide signature --

Scott Barton Kennelly
http://www.bigprintphotos.com

 Scottelly's gear list:Scottelly's gear list
Sony SLT-A65 Sigma SD1 Merrill Nikon D810 Sigma sd Quattro H Sony DT 18-55mm F3.5-5.6 SAM +20 more
D Cox Forum Pro • Posts: 24,632
Re: Chroma Subsampling
1

You can see the chroma in Photoshop. Convert to Lab mode, and fill the L channel with mid grey.

If you blur the a and b channels, you can simulate sub-sampling. With a 2 pixel blur, not much changes.

And with a heavy blur of 16 pixel radius, the picture begins to look hand-coloured.

Scottelly
Scottelly Forum Pro • Posts: 14,217
Re: Chroma Subsampling

Tom Schum wrote:

I used to do TV repair, and once upon a time I also built a Heathkit color TV. I never did the experiment of cutting off the luminance channel, unfortunately.

Color TV was a brilliant compromise in which the color information was added to the existing monochrome TV standard in such a way that the existing base of many millions of monochrome TV sets still worked.

The compromise was based on color perception science, which showed that for some colors the human eye had less acuity than for others. Additionally the target screen size was the large (for then) 25" diagonal 4x3 screen at a distance of 8-10 ft.

This is smaller than a 4x6 photo viewed at 12 inches distance.

I can remember getting up close to the screens of these TV sets (after careful convergence and alignment) and seeing that the edge details always went to gray.

The link you provided was very informative, but it deals with standard color filter array sensor processing. A single "pixel" in such an image is actually an interpolated result of comparing that pixel (with its color filter) to adjacent pixels (with their different color filters). The resolution at pixel level is not equal to the luminance resolution possible, partly because the color filters themselves confuse the luminance information. Both luminance and chrominance have to be derived by interpolation. Single pixel resolution is not possible with interpolation, since more than one pixel is involved in the determination of detail, whether it is chrominance or luminance. Some will dispute this of course.

In the link, single pixel details were mangled but the test was intrinsically flawed by the fact that a color filter array sensor was most likely used to generate the test image.

The only digital imaging system that is able to deliver chrominance resolution equal to luminance resolution is the Sigma Foveon, as seen in classic Foveon and in Merrill, but not necessarily in Quattro, their latest offering. Of course, some will dispute this.

However, such an image might survive the processing quite well.

So do you think a jpeg, created with chroma sampling turned on would represent more closely the maximum image quality capable with a Quattro camera raw file, vs. a jpeg created the same way from a Merrill raw file?

-- hide signature --

Tom Schum
Every day a new image.

-- hide signature --

Scott Barton Kennelly
http://www.bigprintphotos.com

 Scottelly's gear list:Scottelly's gear list
Sony SLT-A65 Sigma SD1 Merrill Nikon D810 Sigma sd Quattro H Sony DT 18-55mm F3.5-5.6 SAM +20 more
xpatUSA
xpatUSA Forum Pro • Posts: 16,037
Re: Chroma Subsampling
1

guzzibreva1 wrote:

http://users.wfu.edu/matthews/misc/jpg_vs_gif/JpgCompTest/JpgChromaSub.html

I remember doing an experiment with the Heath Kit color television I built a long time ago. I disconnected the delay line for the brightness signal leaving only the chroma information to be displayed on the screen. I was amazed at how little color sharpness there was.

Thanks for the link.

We should note that there is no chroma sub-sampling in JPEG images saved from SPP above a certain Quality level (7 or 8?).

See 4:4:4 (meaning NO sub-sampling) here:

https://en.wikipedia.org/wiki/Chroma_subsampling

I may be wrong but I believe that the OOC JPEG 'fine' is sub-sampled 4:2:2 and the X3F's  embedded JPEG (after SD10) is sub-sampled 4:2:0.

The type of sub-sampling can be found in the EXIF tag, for example:

"YCbCrSubSampling: YCbCr4:4:4 (1 1)"

Phil Harvey's ExifToolGUI shows it - DPR's pop-up does not.

-- hide signature --

"What we've got hyah is Failyah to Communicate": 'Cool Hand Luke' 1967.
Ted

 xpatUSA's gear list:xpatUSA's gear list
Sigma DP2 Sigma DP2 Merrill Panasonic Lumix DMC-LX7 Panasonic Lumix DMC-G1 Sigma SD15 +17 more
OP guzzibreva1 Senior Member • Posts: 1,076
Re: Chroma Subsampling

My conclusion from this info is that if you plan to view an image the way you would view it in real live (not magnified by viewing a small portion of the image) you are not bothered by reduced chroma resolution.

Pixel peeping or printing very large and putting nose on the print is when you see the errors.

I have been looking for the reason Foveon images look so good.  It would appear that per pixel color resolution is only part of the reason.

xpatUSA
xpatUSA Forum Pro • Posts: 16,037
Re: Chroma Subsampling

D Cox wrote:

You can see the chroma in Photoshop. Convert to Lab mode, and fill the L channel with mid grey.

If you blur the a and b channels, you can simulate sub-sampling. With a 2 pixel blur, not much changes.

And with a heavy blur of 16 pixel radius, the picture begins to look hand-coloured.

Good illustrations!

For those without Photoshop but still interested, the GIMP can decompose images into Y'CbCr layers - although extracting Cb or Cr in "color" is a bit fiddly (the review layers are displayed in grayscale).

The Cb channel:

The Cr channel:

-- hide signature --

"What we've got hyah is Failyah to Communicate": 'Cool Hand Luke' 1967.
Ted

 xpatUSA's gear list:xpatUSA's gear list
Sigma DP2 Sigma DP2 Merrill Panasonic Lumix DMC-LX7 Panasonic Lumix DMC-G1 Sigma SD15 +17 more
Tom Schum
Tom Schum Veteran Member • Posts: 8,870
Re: Chroma Subsampling

Scottelly wrote:

So do you think a jpeg, created with chroma sampling turned on would represent more closely the maximum image quality capable with a Quattro camera raw file, vs. a jpeg created the same way from a Merrill raw file?

It would make an interesting test, but I don't want to go out on a limb and predict any winners.

A sufficiently high resolution color filter array might deliver better than the Quattro, pixel for pixel, downsized to Quattro 20mp resolution.  Too hard to guess though, without images to look at.

-- hide signature --

Tom Schum
Every day a new image.

 Tom Schum's gear list:Tom Schum's gear list
Fujifilm X30 Sigma dp0 Quattro Fujifilm X-E1 Sigma sd Quattro Fujifilm X-T3 +10 more
Tom Schum
Tom Schum Veteran Member • Posts: 8,870
Re: Chroma Subsampling from PaintShop Pro X9

First, I took a recent lo-res Quattro image, and with SPP I saved a 16 bit TIFF output file. Then, running PaintShop Pro X9 (ver 19.1.0.29 x64), I opened this file and then saved three versions. Then I opened these one by one in Photoshop Elements 13 and cropped out the same part of each image without resizing, and put them into a composite image again without resizing. Any saves in Photoshop Elements were at max quality.

Saved files were, max quality 5133kb, no subsampling high compression 383kb, 4:2:2 subsampling high compression 318kb. This is file size out of PaintShop Pro x9.

These are 100% crops from the files. I have trouble telling the difference between the last two (the highly compressed ones).

PS: using a compression of 90 produced extreme posterization, so I backed off to 80.

PSS: I'm not sure this proves anything...

-- hide signature --

Tom Schum
Every day a new image.

 Tom Schum's gear list:Tom Schum's gear list
Fujifilm X30 Sigma dp0 Quattro Fujifilm X-E1 Sigma sd Quattro Fujifilm X-T3 +10 more
Keyboard shortcuts:
FForum MMy threads