Non-photorealistic rendering of scenes

Abstract

Non-photorealistic rendering (NPR) has been an established field within the area of computer-graphics for several years. The aim of NPR is to create appealing and expressive stylized illustrations from an image, which o↵er a human-drawn representation of a computer-generated render. We note that artistic representation of a photographic scene may be divided into two categories, namely obtaining a sketch-like representation of the scene and obtaining a colourised representation of the scene. While there are several styles of NPR techniques that are described in literature, this dissertation focuses on a painterly style, whereby the image is rendered in such a way that it mimics a painting. In this dissertation, two edge detection algorithms, from which a sketched representation of the scene will be discussed. These are the Gabor filter and an extension to the well known Kuwahara filter. For the colourised representation of a scene, we will be looking into di↵erent ways of adding a textured stroke to an image, both by adding a textured stroke directly on the edges within the image, and also by adding texture to the whole image and enhancing the edges within the image. However, not all artistic styles are the same, therefore we first discuss an overview of current NPR work, and later we focus on four di↵erent forms of renders namely; oil, watercolour, acrylic and pencil sketches. A work of art cannot be evaluated using one simple test due to the subjective nature of the results obtained. Therefore, testing will be divided into three di↵erent sections; visual inspections, quantitative evaluation, and observational studies. The results show that people perceive rendered scenes as a hand-drawn illustration of an image, however for each image a di↵erent approach needs to be considered depending on the elements found within the image.