Interactive Media Systems, TU Vienna

Image Stylization from Stereo Views of Natural Scenes

Thesis by Danijela Markovic

Supervision by Margrit Gelautz and Felix Breitenecker


As media technologies have developed over time, generating novel methods of creating, presenting and interacting with visual information, artists began experimenting with new concepts and possibilities by producing artworks which moved conventional borders of art expression. New media technologies have introduced new opportunities for artistic creation and have brought the computer created design and artwork to a wide spread instrument of visual communication. These recent trends in artistic production and presentation of visual material consequently stimulated the development of novel techniques in the field of computer graphics. Particularly, the (semi-)automatic creation of output that resembles conventional art forms, such as various styles of drawings or paintings produced by hand, has attracted a lot of attention in the computer graphics community. To generate results with a handcrafted-like appearance, a number of systems use a 3D modeled scene and thus benefit from full control over the scene structure. On the other hand, using images of natural scenes in order to compute their stylized representation is a very challenging task. The process of extracting information that is useful for the subsequent stylization from the real input scene can be quite complex and demanding, mostly because of the presence of noise and other artifacts. In spite of this, natural scenes offer a variety of ready-made shapes, objects and textures, which sets them to a powerful and very attractive alternative to 3D modeled scenes for simulating the traditional and exploring new art forms. The focus of this thesis is the stylization of images from real scenes captured using a stereo camera set-up. An essential feature of the depiction process is a novel silhouette extraction approach that we introduce here. This so-called Edge Combination algorithm utilizes stereo-derived disparity maps in order to find silhouettes, which delineate the dominant structure of the scene. Our research results demonstrate that silhouettes extracted using the Edge Combination algorithm are suitable and convenient for image depiction and stylization in various simulations of the traditional artwork, such as sketching and drawing. We also illustrate the benefit of utilizing disparity maps to orient strokes to facilitate the "depth" perception in drawn-like image representations. Finally, we show that silhouettes extracted by the Edge Combination algorithm can serve for motion delineation in imitation of comics produced by hand and also to portray the impression of change in a moving object´s speed.


D. Markovic: "Image Stylization from Stereo Views of Natural Scenes"; Supervisor, Reviewer: M. Gelautz, F. Breitenecker; Institut für Softwaretechnik und Interaktive Systeme, 2007; oral examination: 03-20-2007.


Click into the text area and press Ctrl+A/Ctrl+C or ⌘+A/⌘+C to copy the BibTeX into your clipboard… or download the BibTeX.