Annual Progress Report, Office of Naval Research Grant No. N00014-95-1-1163 (AASERT).
Annual progress rept. 1 Jun 96-31 May 97,
CALIFORNIA UNIV BERKELEY
Pagination or Media Count:
The problems of image sequence compression and new view synthesis have both received a lot of attention recently. In the former case, it is desired to compactly represent the original image set by exploiting redundancy and correlation. This issue is particularly important in applications of storage and transmission. In contrast, the goal of new view synthesis is to generate arbitrary viewpoints of a given scene primarily for visualization purposes. Notice that there exists a trade-off between representation size and the quality of the synthesized images As more views of the scene are added to the representation, the image quality increases as does the representation size. Hence, an interesting problem is to consider both problems at once that is, construct a compact representation which reconstructs the original images and synthesizes new views. We propose two depth-based representations to address these problems. The first approach involves several so called reference frames for which depth and intensity information are both defined. New views are generated by warping the reference intensity and depth data in a manner similar to view interpolation techniques. The second proposed approach integrates all available information with respect to a single reference frame akin to layered representations. The representation then consists of a multivalued array of depth and intensity values which overcomes occlusions and redundancy 4. These depth-based representations both assume the given image sequences arise from a static 3-D scene captured by a moving camera restricted to the x-y plane. Note that the exact motion of the camera is unknown alpha priori and will be estimated.
- Radio Communications