[FRIAM] Kinks in perceptual space
Angel Edward
edward.angel at gmail.com
Sun Mar 5 13:21:55 EST 2023
Steve,
Thanks for the post. I haven’t been following the research in color for a long time although it remains one of the areas I love to talk about in my books and lectures.
The paper is really interesting although I’m not sure how the result could be used in areas such as compression since both small and large differences in color are important. Maybe on the user side but not on the transmission or storage side. The Fast Company article shows a superficial understanding of color spaces at times and thus exaggerates the potential results of the research.
I have a couple of potential issues with the paper that I thought would have been brought up by reviewers. Early the paper, there is statement that seems to say that RGB is what the cones in our eyes sense. Perhaps it's just a poorly stated sentence but the there types of cones do not measure r, G and B in a 1 to 1 way. The three types have peaks in the Green, between Green and Yellow, and in the Blue. In addition, we less than 10% of cones are blue sensitive (and these tend to be on the periphery of the retina). The standard NTSC RGB system was based on the available technology and not on perception. Luminance (L in Lab, Y in XYZ) is mostly Green and Yellow. What makes things more complex is that we are more sensitive to small changes in color (MacAdams ellipsoids of Just Noticeable Differences) in the blue and least sensitive in the Green.
I would have not pointed out the above except that I was worried that the experiment was done totally on luminance, changing L* with a*=b*=0. I would have found the paper more convincing if the measure was applied to changes in chromaticity in the experiment.
We used to worry about these issues way back in the 70’s when I was working with my friend Anil Jain at the USC Image Processing Institute. One problem then finding color maps for compressing a color image form 3 bytes/pixel to 1 byte/pixel and for pseudo coloring grayscale images. The problem was to find a map that in which each step was perceptually equal. Anil did a paper that appeared in the JOSA in 82 using geodesics in a perceptual color space. The problem with the result was that the colors he got were really ugly compared to the usual thermal map using cool colors (blues and greens) going to hot colors (yellows and reds).
Ed
__________
Ed Angel
Founding Director, Art, Research, Technology and Science Laboratory (ARTS Lab)
Professor Emeritus of Computer Science, University of New Mexico
1017 Sierra Pinon
Santa Fe, NM 87501
505-984-0136 (home) edward.angel at gmail.com
505-453-4944 (cell) http://www.cs.unm.edu/~angel
> On Mar 4, 2023, at 4:22 PM, Steve Smith <sasmyth at swcp.com> wrote:
>
> This may or may not relate to the current threads about mind v body, perceptual nature of reality, etc that we have been flogging, but it is a topic I lived in and around for most of my career and found it both familiar/compelling and a little disturbing:
>
> The general topic is the non-Reimannian nature of perceptual color spaces.
>
> The broadly accepted non-Euclidean color spaces described by the CIE formulations (1931 and 1976) has been widely accepted while the RGB/CMYK/HSV Euclidean approximations are what most folks use for pretty good practical reasons (particular the conveniences of tristimulus/process color specification and synthesis).
>
> This recent (1 year old) publication work by some LANL folks was shoved in my face/space recently (as a correlate to the problems we have been working with on trying to understand the underlying space of abstract high dimensional (very non-linear) problems such as ensemble steering/exploration in the World3 model. Our favored method (of the moment) is a variant of tSNE <https://en.wikipedia.org/wiki/T-distributed_stochastic_neighbor_embedding> <https://en.wikipedia.org/wiki/T-distributed_stochastic_neighbor_embedding>which prefers a locally accurate metric over a global one.
>
> The LANL work on this non-Reimannian color space:
>
> https://www.pnas.org/doi/full/10.1073/pnas.2119753119 <https://www.pnas.org/doi/full/10.1073/pnas.2119753119>
> A popular article about that work:
>
> https://www.fastcompany.com/90780869/it-could-take-20-more-years-for-scientists-to-truly-understand-color
> <https://www.fastcompany.com/90780869/it-could-take-20-more-years-for-scientists-to-truly-understand-color>
> I'm guessing this (at least) crossed Ed Angel's awareness, perhaps there are a few others here who care about this level of detail/abstraction on color/perceptual spaces? Frank is probably a lot more up on the nuances of (non) Reimannian manifolds than I ever will be... I don't know if this represents an interesting example of the utility of such?
>
> -. --- - / ...- .- .-.. .. -.. / -- --- .-. ... . / -.-. --- -.. .
> FRIAM Applied Complexity Group listserv
> Fridays 9a-12p Friday St. Johns Cafe / Thursdays 9a-12p Zoom https://bit.ly/virtualfriam
> to (un)subscribe http://redfish.com/mailman/listinfo/friam_redfish.com
> FRIAM-COMIC http://friam-comic.blogspot.com/
> archives: 5/2017 thru present https://redfish.com/pipermail/friam_redfish.com/
> 1/2003 thru 6/2021 http://friam.383.s1.nabble.com/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://redfish.com/pipermail/friam_redfish.com/attachments/20230305/7e8fbd34/attachment.html>
More information about the Friam
mailing list