Hello again! I was asked to do some more analyses for my paper that is still going, years later now, and I got to really investigating my data again. Here is the problem I have found:
My means for the different wavelengths (lw_mean, mw_mean, sw_mean, and dbl_mean) of the clusters identified by naive bayes after QCPA are incredibly low, in the 0.03-0.07 range. That seems like an extremely low value for particularly the long wavelengths (the stripe color is bright red). However, the relative values checked out (long wavelengths greater mean than shorter, etc.).
I am now interested in calculating delta S, so I wanted to see if I could compare the Ssat of my two clusters (stripe and background). I am not quite sure how to do that using the QCPA results, since they report means weighted by the area in pixels the patches take up and I am not interested in weighting by area, but rather a direct comparison. I looked at the other tools and found that there is a Cone Catch to RNL chromaticity tool that should allow me (?) to compare the values, though I still don’t quite know how to do that.
But here is where I noticed a potentially big problem. The saturation values reported with this tool are wildly different than the QCPA results. RNL chromaticity from QCPA was reported to be ~7, while the other tool reported mean saturation of that same area to be 19. So I went further back. I looked at the cluster values measured straight from the cone-catch image (selected my ROIs, press R). The means for lw_mean etc. were much much higher than those reported from QCPA. Looking closer at the two tables, the wavelength means reported in the QCPA table (ROI cluster results) match exactly (all digits) to the standard deviations reported from my measure of the clusters in the cone catch image. So I believe that what is reported in the ROI cluster results from QCPA is actually the standard deviation, not the mean as it is labeled. This caused me to check the Dmax values, which are what I was ultimately interested in getting for the stripe the first round of analysis. Indeed, these standard deviation values are what are used to calculate Dmax (I did the calculations by hand to confirm), not the mean. I do not know if these values are used to calculate other downstream values (VCA, BSA, etc.).
I would love to be told I am wrong, since I am on #4 of redoing this manuscript, so please let me know if you can help me find what I am doing wrong. I also would love some guidance on getting delta S between two patches once all of this gets worked out. Feel free to email me at my UW-La Crosse faculty email address if you don’t want to respond on here. I look forward to hearing from you!
Hi,
I tested the issue on some of my images, as this would be a severe issue. I do indeed seem to get a correct reporting of the mean values of the clusters in a clustered image in the ‘ROI cluster result’ output file. These are not to be confused with the CAA metrics, which account for the relative proportion of each colour pattern element. To calculate pairwise RNL distances between clusters, using either the colour or luminance ‘JND difference calculator’ (Base MICA, i.e., pre QCPA) tools, or manually calculate the distances using the equations in the corresponding literature (Kelber et al. 2003 etc.).
I am not sure which ‘other tool’ you used to compare the cone catch and RNL coordinate output. Please email me a copy of your ImageJ folder and an example image file so I can retrace your workflow and see if I have an explanation for your findings.
Cheers,
Cedric