e-David : Non-Photorealistic Rendering using a Robot and Visual Feedback

Abstract

Non-photorealistic rendering deals with abstraction and representation of information using artistic methods. However, results exist only digitally and not as a physical work of art. In addition, current approaches do not include the process of creation, although it is of decisive importance for the result. The artistic act is dynamic and iterative. Decisions that affect the result are made or adapted during the process. This thesis presents a hardware-based painting system and corresponding algorithms on the basis of a visual feedback loop, which resemble an artist’s technical painting process and generate real paintings with different tools and media.A method is developed that uses the painting robot to create ink drawings. Two styles are combined according to their presentation possibilities, semantic hints and image segments. The process is extended to the creation of colored paintings. The robot applies acrylic paint to a canvas until the resulting painting is sufficiently similar to a given input picture or scene. Two optimization-based algorithms are presented, placing the strokes in relation to existing ones. One in which transparent layers of paint are superimposed on top of each other, and another in which different input colors are approximated by optical blending of palette colors. Another method is presented, which splits input images into regions and layers. The number of strokes and colors needed to display an input image is minimized by this decomposition, which is essential for hardware-based rendering. This method makes it possible to apply different styles of painting to different regions and layers and to create picturesque representations with more artistic freedom. Furthermore, a paint model based on the work of Kubelka and Munk will be presented. The base pigments are selected and their coefficients are determined by a new measuring method that allows to take into account the changes of used paint during the painting process. The paint model also offers the possibility to extract color palettes from images. Furthermore, it can be used to calculate mixing recipes that mix new paint from the base pigments that fits a given target color.