AbstractA technique is presented to constructarbitrary views of a model by using previously computed views. Thetechniqueis simple toimplement, completely general forpolygonal models andcan beused within objecthierarchies or scenegraphs. During a preprocessing step images of a model are taken fromdifferent viewpoints. These images are savedusing long bitfields(ID-bitfields)which encode the visibility information according to an array of the model’sprimitives used as the base for the bitfield.These ID-bitfieldencodings, together with the primitive array, are thenused by a viewer which selects and joins them to provide an approximated (not conservative) reconstruction ofthe visible elements of the object for a new viewpoint. The technique implicitly performs occlusion culling, since aminimal set of visible polygons is the result of the reconstruction. Results show how interaction can be improvedwhen working with high depth complexity models. Satisfactory reconstructions are achieved by taking as few as25 images around an object. This paper suggests how the technique can be extended to other applications such asvirtual walkthroughs and visualization of non-realistic images, and how graphics libraries and hardware could beenhanced by allowing the application to pass an ID-bitfield.Key words: view reconstruction, interactive display, visibility preprocessing, occlusion culling, polygon reduction.1. IntroductionCurrent graphics systems typically solve the problem of de-termining the set of visible polygons for a viewing point andviewing direction (definedby the camera position and orien-tation) using the z-buffer algorithm. During rendering, the z-buffer algorithm resolves the visibility problem by compar-ing the depth information of new primitives sent to the ren-dering buffer with the depth information of the z-buffer ona pixel-per-pixel basis. The z-buffer algorithm is relativelyeasy to implement and is widely available in hardware forcurrent graphics systems. However, it is not always efficientto solve visibility using this algorithm, since the algorithmneeds to be applied for every frame rendered. Large modelscontaining thousands of polygons or several layers that oc-clude each other (e.g. buildings with several rooms or the fo-liage of trees) are not good candidates for using the z-bufferalgorithm, because the number of polygons occluded by theset of visible polygons may be extremely large, consumingsignificantamounts of time during rendering.
[1]
Richard Szeliski,et al.
Layered depth images
,
1998,
SIGGRAPH.
[2]
David Salesin,et al.
Hierarchical image caching for accelerated walkthroughs of complex environments
,
1996,
SIGGRAPH.
[3]
Ken Shoemake,et al.
Arcball Rotation Control
,
1994,
Graphics Gems.
[4]
Dinesh Manocha,et al.
Visibility culling using hierarchical occlusion maps
,
1997,
SIGGRAPH.
[5]
Takafumi Saito,et al.
Comprehensible rendering of 3-D shapes
,
1990,
SIGGRAPH.
[6]
Amitabh Varshney,et al.
Navigating static environments using image-space simplification and morphing
,
1997,
SI3D.
[7]
Seth J. Teller,et al.
Real-time occlusion culling for models with large occluders
,
1997,
SI3D.
[8]
Oliver Deussen.
Pixel-Oriented Rendering of Line Drawings
,
1997
.
[9]
Joëlle Thollot,et al.
Conservative visibility preprocessing using extended projections
,
2000,
SIGGRAPH.
[10]
Thomas Strothotte,et al.
Computational visualization - graphics, abstraction, and interactivity
,
2011
.