![\"20161202_094258-2\"](\"http:\/\/barradeau.com\/blog\/wp-content\/uploads\/2016\/12\/20161202_094258-2.jpg\")
<\/a>Limestone figure wearing a mask<\/a><\/p><\/div>\n Limestone figure with mask<\/a><\/p><\/div>\n limestone statuette of herakles brandishing a club<\/a><\/p><\/div>\n limestone statuette of eros<\/a><\/p><\/div>\n all the images are based on photos from Archive.org. I wanted to find a way to generate non photorealistic versions of these pictures,\u00a0I started describing the process\u00a0here<\/a>\u00a0;\u00a0in a nutshell:<\/p>\n extract the saliency > dithering (Floyd Steinberg) > reaction diffusion > component labelling & vectorisation.<\/em><\/p><\/blockquote>\n the images are\u00a0pre-processed through Ubernet, computed in javascript then\u00a0post-processed with Potrace and illustator.<\/p>\n my plan was to use the visual salience<\/a>\u00a0 of the images to drive the generation and distribution of points then try to find something different from the (overused) delaunay triangulation \/ voronoi diagram.\u00a0first I used MatLab to compute a saliency map<\/a>\u00a0using this add-on<\/a>.<\/p>\n matlab saliency map<\/p><\/div>\n it worked\u00a0; the yellow areas of the saliency map are the “most interesting parts” of the picture, distributing more points there is trivial but the map was a bit too far from the actuual morphology of the object ; here for instance it would give a great weight to the neck, the beak and the eye completely ignoring the cockscomb (funny word).<\/p>\n then I vaguely remembered Ubernet<\/a>\u00a0by Iasonas Kokkinos<\/a> (pdf<\/a>), that produces this kind of images:<\/p>\n Ubernet indicative result<\/p><\/div>\n Ubernet\u00a0is a<\/p>\n `Universal’ Convolutional Neural Network for Low-, Mid-, and High-Level Vision using Diverse Datasets and Limited Memory<\/em>.<\/p><\/blockquote>\n which I read\u00a0bluntly as<\/p>\n a magic tool that does exactly what\u00a0you want.<\/em><\/p><\/blockquote>\n now just look at the\u00a0saliency map! that’s exactly what I was after and the code was made available as an online API\u00a0http:\/\/cvn.ecp.fr\/ubernet\/<\/a>\u00a0:)<\/p>\n here’s the same picture’s saliency map computed with Ubernet:<\/p>\n this was\u00a0better\u00a0suited to my needs, I used the online API to spit saliency maps\u00a0& edges maps (and normal maps… just in case… and because normal maps are\u00a0awesome ^^).<\/p>\n the second step was to perform a dithering<\/a> to obtain an interesting points distribution\u00a0;\u00a0dithering results in a binary (black and white) image which is easier to vectorize.<\/p>\n that’s where I understood that the most interesting patterns to vectorize appeared in the “grey areas” rather than in the highly contrasted ones.<\/p>\n I later skipped the dithering in favor of a Reaction Diffusion pattern, the RD lets this kind of “grey areas” patterns emerge\u00a0and\u00a0– as the RD patterns emerge\u00a0– the dithering is completely obliterated. RD\u00a0produces the image below after a couple of iterations.<\/p>\n Reaction Diffusion can be achieved in many ways, the fastest \/ dirtiest I know of is to successively apply a sharpen filter (reaction) then a blur filter (diffusion), as simple as that. the good thing is that both are convolution fiilters<\/a>\u00a0(the examples don’t work but the snippets will).\u00a0First I implemented the filters manually but it was very slow and hard to fine tune so I used glfx.js<\/a>\u00a0to speed up the process and gain more control over the parameters.<\/p>\n below is a series of settings that show how the result varies depending on\u00a0the radius of the sharpen\/blur filters ( r: 0.5, 1, 2, 3, 4, intensity: 1.5 ).<\/p>\n the first row show the 3 input images: the source, the edges and the salience map computed by Ubernet,\u00a0the second row shows an unconstrained<\/em> Reaction Diffusion pattern emerging and the third row shows the same pattern constrained by the saliency and edges map.<\/p>\n<\/a><\/a><\/a>
\n<\/a>I\u00a0didn’t have anything special in mind, I just found these objects moving\u00a0; they were like action figures from long ago. the statues are quite small too, the printed version is roughly their actual size, small things are cute.<\/p>\nSaliency Map<\/h2>\n
<\/a><\/p>\nDithering<\/h2>\n
<\/p>\nReaction Diffusion<\/h2>\n
<\/p>\n