Open Collections

Post-processed shadow determination for composition of depth images

University of British Columbia

In the creation of computer generated pictures, many methods have been proposed to reduce the amount of time it takes to render an image or sequence of images. Efforts have been made to improve existing rendering algorithms and to create better ones, but the compositing method has received little attention. The compositing method breaks a scene up into parts which can be rendered separately which are then put together in such away that the result is nearly indistinguishable from an image rendered from the complete scene. This can not only reduce the total time taken to generate an image sequence, but can also create composite images from others rendered by different methods. In 1984, Porter and Duff first introduced a well defined approach to computer image composition, and in 1985, Duff proposed an extension to the this method that employed depth images. Using depth images, visibility determination can be done by the computer, reducing the human effort required to produce the resultant image. The images produced are likely to be missing some elements that would be present if all objects were rendered in a single scene, such as shadows from objects in one image not being cast on objects from other images. This thesis addresses the problem of modifying composited images such that the resultant picture contains shadows that would normally only be present in an image rendered from a scene containing all objects in both original pictures. A simple algorithm is developed to reconstruct the visible surface of objects in an image. The reconstruction is used to generate a shadow map for each image to be composited. The shadow map indicates where shadows would fall on one image were objects in the other image present when the original scene was rendered. The shadow map is then used to create shadows in appropriate places on the original images, and these reshadowed images are composite to obtain the desired result. The application of these techniques to other areas is also investigated.

Thesis/Dissertation

application/pdf

2008-09-16

For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use.

http://hdl.handle.net/2429/2054

Computer Science

University of British Columbia

UBCV

DSpace