THESIS
2001
ix, 62 leaves : ill. ; 30 cm + 1 computer optical disc (4 3/4 in.)
Abstract
Traditionally, making three-dimensional animations of a scene depicted in single image is dfficult even for skilled animators. Computer vision techniques cannot be applied to automatically extract three-dimensional models from a single image; thus tedious trial-and-error manual approaches are needed to build models that closely resemble the scene depicted in the image....[
Read more ]
Traditionally, making three-dimensional animations of a scene depicted in single image is dfficult even for skilled animators. Computer vision techniques cannot be applied to automatically extract three-dimensional models from a single image; thus tedious trial-and-error manual approaches are needed to build models that closely resemble the scene depicted in the image.
Image-based modeling and rendering (IBMR) has emerged as an attractive means for producing stunningly realistic novel views of scenes from a set of reference images, without first constructing the geometric models. In this thesis, we introduce two IBMR approaches for making fly-through or walk-through animations from a single large landscape painting or panorama. These images can either be regarded as having sub-scenes taken from different perspective or as exhibiting a very wide view. The first approach constructs a simple global model for the entire input image and builds a local model for each sub-scene using the TIP spidery mesh interface. An animation is generated by switching smoothly between a local model and the global model while moving a virtual camera along an animation path. Novel views are rendered by mapping the texture images, which are extracted from the original image, onto the active model. The second technique assumes that the image is taken from a fixed viewpoint and constructs a unified model for the entire scene, thus eliminating the need to perform model switching. In addition, we propose a minor improvement on the TIP spidery mesh user interface to make better use of visual clues presented in the image while modeling the background scene.
Post a Comment