Ad blocker detected: Our website is made possible by displaying online advertisements to our visitors. Please consider supporting us by disabling your ad blocker on our website.
Does anyone have insight into how PC Stitch, KG Chart, etc. does color quantization?
I can run 3 different picture to patter programs and get 3 different patterns. I assume that somewhere they are using an RGB color assigned to each thread and doing a color difference (http://en.wikipedia.org/wiki/Color_difference) to determine the "closest" thread color. For some time now, I have been wondering if the RGB values they have assigned to the threads are skewed lighter than what they really are. I have hypothesized whenever a color swatch was stitched of that thread, it was done on white aida cloth (14 count) which causes the white to seep into the average, giving the thread a brighter RGB than what it really is. I exclusively use 18 count aida with 2 threads. I rarely have problems with the background seeping through unless I am working with very dark colors (especially black). This has become a reoccurring problem and has caused me lots of headaches.
I was wondering about this too. I have been considering creating a program for cross stitch creation that has the ability to use machine learning to "correct" color choices by using user input. Would anyone be interested in that?
I doubt that colour swatches are stitched. I would assume that the colours are simply based on the threads themselves. There are a few different resources online that state that they have the RGB values of various threads but they are never perfect. I always check the colours generated by a program before stitching. It takes a long time but it is less frustrating than working on a piece and then eventually realizing that the colours don't work. The natural sheen on the threads may also affect the results of a machine trying to determine RGB values for floss.
Here's an example of the difference between what you see in KG-chart and the actual piece. (the colours weren't actually picked by KG-chart, I just inputted them manually after selecting what I liked)
Spoiler
chrono comparison.png (535.85 KiB) Viewed 6677 times
Current Projects:
- Game of Thrones/Mario World Mash-up (cross-stitch)
RoseLight wrote:I was wondering about this too. I have been considering creating a program for cross stitch creation that has the ability to use machine learning to "correct" color choices by using user input. Would anyone be interested in that?
I think that the variations in dye lots is probably a good reason that programs don't come up with the colors we like, in addition to the bit where computers display things differently than eyes do. I know I've seen multiple people mention that they had multiple skeins of the same color number, but they looked quite different because the dye lots were different.
Mishatu wrote:I think that the variations in dye lots is probably a good reason that programs don't come up with the colors we like, in addition to the bit where computers display things differently than eyes do. I know I've seen multiple people mention that they had multiple skeins of the same color number, but they looked quite different because the dye lots were different.
This happened to me before. They were so different in fact that I ended up having to contact DMC directly to try to find another of the same dye lot, which didn't end well.
I also have plans (in my ample free time... which hasn't existed recently) to finish out my own pattern software, including more accurate colors. What I've done so far with it is mostly disappointing, because different peoples screen representations of the color can vary tremendously. I still intend to do it, though, if someone doesn't beat me to it! [hint: someone beat me to it please because i probably will never have time. ]
I was surprised to see this thread got some attention. The way I see it there are two parts to this problem. 1. The matching of a color to a thread. 2. The trust in that match.
Here is my proposed solution:
1. Make 10 x 10 color swatches for each of the DMC threads.
2. Scan them in and get their average RGB values
3. Write a color matching program.
I really don't think that scanning will give an accurate result.
I was thinking that I could take the current values and use them and then as people correct colors I could correct based on that. So I would generate based on what people have previously corrected. This way the software would become better and better without as much of an upfront correction cost.
1. Make 10 x 10 color swatches for each of the DMC threads.
2. Scan them in and get their average RGB values
3. Write a color matching program.
This still has the issue that two skeins of the "same" color can look drastically different. Especially if bought on different continents. And how you would get your swatch of the multiple variations would be difficult to work out as some colors will have more difference than others. Also, how do you correct for someone who feeds you a color that isn't true, or who has their computer settings odd which means it really doesn't match for them, whereas it might for others? So many variables, I don't honestly see how you can get this to work in all cases.
]