bulia byak wrote:
Questions:
- Within "multiple scanning", what's the difference between
"brightness" and "monochrome"? They give different previews but very similar results. Besides "brightness" is confusing with the brightness method. I think we should leave only one.
These are quite different, actually. Brightness has the same algorithm as the Brightness Threshold above, but instead of merely diving the image into 2 light and dark regions, it partitions the brightness space into however many pieces the user wants. It merely adds r,g,and b to get the brightness, and throws color information away. It is good for areas of different brightness, but fails for areas of similar brightness but different color.
Color does a color quantization, and scans for each color.
Mono does the same as color, and merely removes the hue from the indexed palette after quantization. It is like a convenience tool, so that a user who wants a grayscale set of paths does not need to use 'Color' then the Fill-stroke tool to remove the colors. It is quite different from brightness, actually. They do look similar from a distance, but very different up close. It is like the difference between an old b&w movie, and a color movie showing a scene in b&w. You can tell the difference.
- Import this image (b/w jpeg):
http://www.livejournal.com/userpic/569297/360079
and trace it: multiscan, monochrome, 8 colors. Results are quite good. Now do it again at 9 colors - the result becomes much worse and flatter, as if you reduced the number of colors, not increased it. Then try 10, 11, etc colors: the color resolution does not improve, but the trace becomes darker and darker. After 12 it brightens up again and later starts behaving unpredictably. At 16 colors, each click on Preview gives a different preview image (does not happen at 8 colors). Doing the same in color is similar: the best result is at 8 colors, then starts darkening and degradation, then some wild colors appear. I think this is some bug in the algorithm.
I think that the darkness is caused by the default transparency of 75%, making the deeper multi-traces darker. I'll try "fill-opacity:1.0", as suggested by mental.
I'll look for any bugs you mentioned. It should be threadsafe. There is no static data. Intermediate data is all thrown away.
Also, keep in mind that octree reductions might not have satisfactory solutions for some color subsets. To get the number of colors desired, a complete octree reduction is performed, and then it is pruned to the best N colors. Then the pixels in the image are each adusted to the closest color in the palette. So each N set of colors can have a very different result.
Things to do:
- Since "color" and "monochrome" are in the same radio button group as
all the other scanning methods, they should be stacked vertically to align with them.
I'd like to not waste too much space, and I wanted to make it clear that they are all similar. But how about this? Instead of: ()Brightness (o)Color ()Mono Colors:[8.0] [x]Smooth [x]Stack
... how about this?
()Brightness [x]Smooth (o)Color [x]Stack ()Mono Colors:[8.0]
- Please add tooltips to all widgets in the dialog.
Okay. Any easter eggs? ;-)
- Please fill in the release note on this feature
- Please prepare a large detailed screenshot with comments, to be
added to the site.
I'll try tomorrow, when I can get to a monitor as large as the ones you guys apparently have. ^^
Bob