have seen in
previous edition at V2_ and at DEAF2012 the 100% democratic
Wants to Be…? is all about voting. V2_ developer Artm Baguinski created a Vote-counter app for The People Speak, he demonstrates it
in this video.
How it works? Very easy, Artm Says: "When counting, all pixels of the incoming picture (converted to CIE Lab color space) are classified using K Nearest Neighbors search with K=1. The algorithm builds two maps: indices of the most-similar color per pixel and dissimilarities between pixel color and chosen palette color. The dissimilarity image is then thresholded on a value that user can interactively adjust. While finetuning the threshold value user sees the result of the thresholding as a posterized version of the input image with the pixels too dissimilar to one of the learned card colors painted black. After thresholding the dissimilarity map is split into three, one for each card color. Contiguous contours are searched in each of them and are shown as white outlines on top of the original image. Not all contours are shown / counted though - additional contour-area filter selects only blobs that are larger than a second interactively found threshold."
Have fun and keep voting!
Vote counter manual from Артём Багинский on Vimeo.