《Supervised Learning of Edges and Object Boundaries》.pdf

Supervised Learning of Edges and Object Boundaries ´ ∗ ∗ Piotr Dollar Zhuowen Tu Serge Belongie Computer Science Engineering Lab of Neuro Imaging Computer Science Engineering University of California, San Diego University of California, Los Angeles University of California, San Diego pdollar@ zhuowen.tu@ sjb@ Abstract high level information (e.g. object knowledge [19]). Canny also cannot take into account local information at multiple Edge detection is one of the most studied problems in scales. Such information is important: sometimes we hal- computer vision, yet it remains a very challenging task. It is lucinate a boundary where there is weak or even no local difficult since often the decision for an edge cannot be made evidence (e.g. certain parts of an object may have the same purely based on low level cues such as gradient, instead we intensity pattern as the background), other times we do not need to engage all levels of information, low, middle, and see a boundary even if there are strong local cues that would high, in order to decide where to put edges. In this paper imply its existence (e.g. in the presence of shadows). Com- we propose a novel supervised learning algorithm for edge plex generative models, such as presented in [19, 15], have an