Xu, XuemiaoZhong, LinyuanXie, MinshanQin, JingChen, YilanJin, QiangWong, Tien-TsinHan, GuoqiangDavid Mould and Pierre Bénard2015-06-222015-06-222015https://doi.org/10.2312/exp.20151191We present a fast structure-based ASCII art generation method that accepts arbitrary images (real photograph or hand-drawing) as input. Our method supports not only fixed width fonts, but also the visually more pleasant and computationally more challenging proportional fonts, which allows us to represent challenging images with a variety of structures by characters. We take human perception into account and develop a novel feature extraction scheme based on a multi-orientation phase congruency model. Different from most existing contour detection methods, our scheme does not attempt to remove textures as much as possible. Instead, it aims at faithfully capturing visually sensitive features, including both main contours and textural structures, while suppressing visually insensitive features, such as minor texture elements and noise. Together with a deformation-tolerant image similarity metric, we can generate lively and meaningful ASCII art, even when the choices of character shapes and placement are very limited. A dynamic programming based optimization is proposed to simultaneously determine the optimal proportional-font characters for matching and their optimal placement. Experimental results show that our results outperform state-of-the-art methods in term of visual quality.I.3.m [Computer Graphics]MiscellaneousVisual Art10.2312/exp.20151191183-193