Since the mid 1980s, Non-Photorealistic Rendering (NPR) has gradually become one of the hottest focus in computer vision and graphics. Different from the conditional photorealistic rendering, NPR focuses on the various styles of artistic features and effects of images instead of the truthfulness of physical rendering. Digital art painting also plays an important role in the research of NPR. In the research of painting, the core process is to generate a personalized artistic painting from an image using several interactions. In order to make the effect closer to the artistic effect finally, we develop some algorithms combine the computer vision techniques and the graphics approaches to simulate three kinds of style, Sketching: We propose a novel approach to extract sketch from the natural image using low-, mid-level cues. Some sketch properties, including edge scale, edge value etc, are also computed during the extraction period, which are used to control the sketch rendering. Embroiderer: In order to place the embroiderer thread, we build a layered orientation field through the image segmentation, primal sketch computation and orientation diffusion. Then we simulate the embroiderer thread with a texture-mapping approach. At last we place the embroiderer thread following the orientation field. Oil-Painting: We build a system to generate oil-painting style using interactive image parsing. In the image parsing stage, we extract the image information, such as image interactive segmentation and automatically object classification, sketch extraction, orientation field computing, from the input image. In the rendering stage, we build a brush dictionary and apply different types of brush for different objects. To simulate better oil-painting effects, we use a new approach which can transfer the input image to the oil-painting style from the color view.
修改评论