Ya, I think that they just do some sort of either color averaging or just picking the top 3 or 4 colors. They do seem to do some sort of analysis for shapes – probably some type of edge detection.
Mr Martian’s post above has a lot of circles
And the Rothko one has lots of lines.
The edge detection filters tend to depend on contrast, so your two pics might not work quite as well.
The randomness in composition might be just randomly picking a “direction” for the filter.
I tried to repeat it and noticed that even repeating the original input produces unique output every time.
Might be. Normally, though, it doesn’t usually take that great a contrast to edge-detect, more like a well-demarcated contrast, i.e., sharp edges, little bleeding on either side of the edge, and both Parliament of Fossils (the grey pic) and Composition XIX have that to a greater degree than the Rothko. I’m saying this advisedly - these are digital pics, and my working methods use a lot of edge detection.
I suspect the algorithm starts with colour picking and averaging, and that is where these pics are losing it. I say colour averaging because that brick colour isn’t actually to be found to a large extent in Composition XIX - just mainly around the left edge - but there are a lot of colours in use that are nearby on the colour wheel (reddish blues and purples, reddish oranges).
I think they may be working with a hue/luminosity/saturation model of colour, because Composition XIX, our worst case, has large contrasts of hue, modest contrasts of luminosity, and very little contrast in saturation. I suspect that the LS components are overwhelming the H components when it comes to picking, and the algorithm is thus averaging across the entire picture. When that’s done, there is nothing to edge detect.
Yup, I had noticed that. They’ve got a randomiser going in the later stages of processing.
Interesting. But a little TOO simple for my tastes.
This topic was automatically closed after 5 days. New replies are no longer allowed.
