5 days later

abedhaddad We are on it at last.

At the moment there is an issue with the averaging approach. The aging curves are not measured at exactly the same points in time. I mean, after you start an aging, the time intervals from one measurement to the next does not be the same in different curves.
The jitter comes from the operating system itself.
How would you like us to make an average? If we have an aging curve A and we use its time coordinates for averaging,
we would have to interpolate all the other curves at those coordinates.
However, interpolation is kind of averaging already, right? It takes two points and finds a third one in between. And you can treat it as weighted average of the two points.
So, if we have curves B, C, and D, and we average them with A than there is a mathematical bias right there.
In a sense, B, C, and D are “more” averaged than A.

You may think that, for example, taking first average of all the time points (from A, B, C and D) and do interpolation on all of the curves does not remove the bias from the data.

      zbyszek

      @zbyszek

      Hello,

      I see what you mean and I did encounter this problem. So how about 1) fitting each point to a cubic spline and then 2) interpolating the data to a specified time interval that is customizable (every 1 second, 10 seconds, 30 seconds, 60 seconds) to have uniform time intervals across all curves and then 3) averaging the resulting curves to one curve with standard deviation at every resulting point?

        2 months later
        7 days later
        a month later

        Here is a glimpse of the the new functionality of the MFT software

        In the Aging tab there is a sub-tab “Analytics” which will let you compose averages.

        The plot above shows a single average with a Standard Deviation indicated by the semi-transparent shadow.

        And here are the components for the average as well …

        Hi, this looks great! Happy to see this progress- much appreciated for sharing on this forum.

        Will you be able to also average other kinds of data on the Y axis, like DL, a, b, etc..? Do you have an idea of when this version will be rolled out?

        Thanks!
        Abed

          abedhaddad
          We were told that what conservation people are interested in are the color change aggregates like ΔE76, so to keep things simple we have included only those in the first release candidate (rc).

          If everything goes as we expect the first rc will be available within two days. It will have averaging of the plots, averages in the pdf reports but no averages exported into csv files yet.

            First release candidate with averaging is available from: here.
            Should be intuitive …

            Edit 29/07/2022:
            New version of the software added. The changes include:

            • improved stability
            • “Remove” button added so that averages can be dismissed
            • Better graphics and plot scaling with STDs.
            • Automatic switch to “Spectrum” tab after auto-focus
            12 days later

            @zbyszek

            This is great news!! Thanks you. So will the future iterations have averaging for DL, a, b, etc… those are just as crucial from the conservation science perspective.

            Also the download is not working for me from the ownCloud. could you or @Piotr share via email? Thanks !
            Abed

              abedhaddad

              Really?
              You are the second person with this request. Apart from perhaps ΔL the coordinates of the L*a*b space don’t have physical meaning. And to my knowledge they are not used directly at all, are they?
              ΔEs do have very practical interpretation for conservators. They represent overall color change.

              ΔEs are functions of L, a and b.
              As a general rule, it is risky (euphemism) to average arguments of a function rather than the function itself.
              The only exception is when the function is linear, so it doesn’t matter where you do the averaging.
              ΔEs a non-linear.
              Ergo, adding averages of L, a and b would only mislead MFT users and encourage them to do unorthodox (another) things.

              Of course, I could be wrong. Could you provide an argument supporting the need for L*a*b coordinates averaging?

              Which browser are you using for downloading the software? Could you use chrome or firefox?

                abedhaddad

                Hello Abed,
                Third time’s a charm. I have been dumped from the page twice already. Ok. When I work with MFT data from different regions of interest (ROI) and I want to determine if the ROI have a similar composition, I first look at the spectra, but then I compare the change (d) of the CIE L,a,b vales for each ROI. If the spectra are similar and the dL, da and db are in the same direction, and similar magnitude, rate and shape respectively then I group the ROI. To this end, a feature that calculates the average of the dL, da and db with STD error bars would be useful.

                I will post this and then continue.
                j

                  abedhaddad

                  I keep getting dumped. Ok, again. When I group ROI, I still calculate the dE00 of each ROI from the raw (unaveraged, but perhaps smoothed) CIE L,a,b values. Then I calculate the average dE of the group. I have tried to aggregate the data from ROI and calculated the dE00 from averaged CIE L,a,b values. This aggregated dE00 of the grouped ROI is often different from the averaged dE00 of the grouped ROIs. This makes me nervous, and I am not enough of a colour scientist or statistician to make sense of it and reason why. Perhaps we could get Roy Burns to weigh in on this?

                  j

                    abedhaddad

                    In conclusion, yes please to average of dL, da and db. I do not think that averages of dh or dC would be all that useful, but i am open to be corrected.

                    j

                        Hi @zbyszek and @jacob-thomas

                        I agree with @jacob-thomas that it is necessary to average DE and the lab values (and DL, Da, and Db). We always use those values not only to compare different spots from an ROI as @jacob-thomas suggests, but also to describe the change (ie, getting darker, yellower, cooler, etc). This is very helpful for conservators and conservation scientists alike so I think it is paramount that we do that. I also like to average DC because it also identifies saturation so averaging that would be of interest to me. The algorithm averages the DE curves and does not rely on an average of the each of the variable first, I understand you correctly, @zbyszek . So it should be okay to look at individual averages to identify descriptive trends.

                        And @jacob-thomas as for the average of an aggregate, I try to stay away from that because of the many factors that could influence a DE00 value, such as texture, gloss, pigment concentration, etc… So, this is when I do turn to the more qualitative descriptions to discuss areas of similar chemical composition and look to the DL, Da, and Db values a bit more than DE00 itself. I only average areas that are in close proximity to each other in a single ROI rather than all over an object of interest. Also is your aggregate within the SD of the ROI average (and vice versa?) I know Julio and Mario have looked at a mathematical model to correct for the difference between runs. See Mecklenburg MF, del Hoyo-Meléndez JM. Development and application of a mathematical model to explain fading rate inconsistencies observed in light-sensitive materials: Mathematical model to explain fading rate inconsistencies in light-sensitive materials. Coloration Technology. 2012;128:139–46.

                        Also @zbyszek I am using chrome and I couldn’t download. Thanks to both of y’all!

                          abedhaddad and jacob-thomas

                          Now I get it. So, we are not asked to compute ΔE from averaged ΔL, Δa and Δb.
                          You just use of those averages for other things.

                          We will add the averages of ΔL, Δa and Δb to the software.

                          Are the Δh and Δc also required? It will get a bit crowded in the analytics tab which by itself is detrimental to the user-frendliness of the interface.

                              @zbyszek and @jacob-thomas

                              okay perfect! I’m glad that we figured it out 🙂

                              My feeling is to have all of the averages because you never know who will want them now or in the future for very specific studies. I vote for having them- I especially like DC in particular. There are already a bunch of options in the Aging Curves interface so I don’t think it will be any different really from a user friendliness perspective.

                              Thank you!!

                              abedhaddad

                              I am using chrome and I couldn’t download. Thanks to both of y’all!

                              For chrome just copy the download link to the address bar of the browser and press enter …

                                  zbyszek that worked! Thanks- i’ll be looking forward to the the rest of the averages 🙂

                                  Abed