Figure 1 . Examples of creations realized with DataQuilt, a new interactive authoring tool that allows authors to borrow visual and stylistic elements from raster images and re-purpose them to create custom, pictorial visualizations. Left: a scatterplot of famous paintings by Klimt, showing the date of creation (x-axis) against how much it was sold for in auction (y-axis). Each data point is a spiral-shaped glyph whose texture is mapped to the painting it represents, whereas the Tree of Life

more »

... nting is used as a decorative background. Middle: a bar chart representing the distance from the sun for the planets of our solar system. Each data point is represented by a space rocket, whose exhaust flames are stretched according to the underlying data. Decorative glyphs (sun, planets) are used for further information and visual appeal. Right: A personal visualization depicting one's coffee intake over a week. The type of coffee (espresso, latte, etc.) is represented by different coffee cups, all extracted from photographs. The orientation of the handle represents the time, whereas size is proportional to the drink size and horizontal position corresponds to the day of the week. ABSTRACT Recent years have seen an increasing interest in the authoring and crafting of personal visualizations. Mainstream data analysis and authoring tools lack the flexibility for customization and personalization, whereas tools from the research community either require creativity and drawing skills, or are limited to simple vector graphics. We present DataQuilt, a novel system that enables visualization authors to iteratively design pictorial visualizations as collages. Real images (e.g., paintings, photographs, sketches) act as both inspiration and as a resource of visual elements that can be mapped to data. The creative pipeline involves the semi-guided extraction of relevant elements of an image (arbitrary regions, regular shapes, color palettes, textures) aided by computer vision techniques; the binding of these graphical elements and their features to data in order to create meaningful visualizations; and the iterative refinement of both features and visualizations through direct manipulation. We demonstrate the usability of DataQuilt in a controlled study and its expressiveness through a collection of authored visualizations from a second open-ended study.