Computer graphics processing and selective visual display system – Computer graphics processing – Three-dimension
Reexamination Certificate
2002-06-18
2004-08-17
Zimmerman, Mark (Department: 2671)
Computer graphics processing and selective visual display system
Computer graphics processing
Three-dimension
Reexamination Certificate
active
06778173
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to computer graphics and more specifically to a three-dimensional still and animated object representation obtained from photos of real objects and their geometrical representations, and to a method and apparatus for representation and rendering, using Binary Volumetric Octree.
2. Description of the Related Art
In the immediate future, high-quality rendering of three-dimensional (3D) objects at interactive speed will receive the primary emphasis in modern graphic systems. The demand for high-quality rendering of 3D objects necessitates effective algorithms to be devised for compression of the objects and transmitting them via communications networks in such fields as electronic commerce, computer games, science, engineering, medicine. Use of traditional polygonal models of 3D objects during the last tens of years to simultaneously meet all these demands has failed to give the desired result. Polygonal models have two major shortcomings: large volume (e.g., realistic models require tens of million triangles) and difficulty of constructing.
To overcome these difficulties, several approaches to 3D graphics were suggested in recent years. The most advantageous of them seem to be methods based on using images of objects, and methods based on using points instead of triangles in 3D space.
Image-based methods represent the given object as a set of images—‘photos’ of the object—totally covering its visible surface, and taken from several different camera positions. Besides, each such image is accompanied with corresponding depth-map which is an array of distances from the pixels in the image plane to the object surface. An advantage of such a representation is that reference images can provide high quality of the object visualization regardless of its polygonal model complexity, and can be compressed by usual image compression techniques without sacrificing much quality. In addition, rendering time is proportional to the number of pixels in the reference and output images and not the object complexity.
Disadvantages are due to the fact that obtaining depth maps for real life objects (e.g., sculptures) is rather complicated operation, as well as to insufficiently developed techniques of handling such representations.
Point-based methods represent an object as a ‘cloud of points’ without imposing explicit local polygonal structure. In this method, a set of images with a depth defines a set of points (having corresponding colors) on the object surface by translating each pixel of each reference image by the corresponding depth value in the direction orthogonal to the image plane. Hence image-based representations are a particular case of point-based representations. In the following we shall concentrate on image-based representations as they are closer to our approach.
In literature, the two aforementioned trends are described in references [1] to [13] describing such 3D object representation and rendering methods, as Relief Textures Mapping [1], Layered Depth Images [2], Layered Depth Image Tree [3], Qsplat [4], Surfels [5] and some other that have been known in prior art. In the following discussion of the prior art approaches, references will be made to the following publications:
[1] Manuel M. Oliveira, Gary Bishop, David McAllister. Relief Textures Mapping, Proceedings of SIGGRAPH '00;
[2] Jonathan Shade, Steven Gortler, Li-wei He, Richard Szeliski, Layered Depth Images, Proceedings of SIGGRAPH '98;
[3] Chun-Fa Chang, Gary Bishop, Anselmo Lastra. LDI Tree: A Hierarchical Representation for Image-Based Rendering, Proceedings of SIGGRAPH '99;
[4] Szymon Rusinkiewicz, Marc Levoy. QSplat: A Multiresolution Point Rendering System for Large Meshes, Proceedings of SIGGRAPH '00;
[5] Hanspeter Pfister, Matthias Zwicker, Jeroen van Baar, Markus Gross. Surfels: Surface Elements as Rendering Primitives, Proceedings of SIGGRAPH '00;
[6] Chamberlain et al., Fast Rendering of Complex Environments Using a Spatial Hierarchy, Proceedings of Graphics Interface '96;
[7] Grossman and Dally, Point sample rendering, Proceedings of Eurographics Workshops on Rendering Techniques '98;
[8] Lischinski and Rappoport, Image-Based Rendering for Non-Diffuse Synthetic Scenes, Proceedings of Eurographics Workshops on Rendering Techinques '98;
[9] M. Levoy and T. Whitted, The Use of Points as Display Primitives. Technical Report TR 85-022, The University of North Carolina at Chapel Hill, Department of Computer Science, 1985;
[10] L. Westover, Footprint Evaluation for Volume Rendering, Proceedings of SIGGRAPH '90;
[11] C. I. Connolly. Cumulative Generation of Octree Models from Range Data, Proceedings of Intl. Conf. Robotics, pp. 25-32, March 1984;
[12] G. H Tarbox and S. N. Gottschlich. IVIS: An Integrated Volumetric Inspection System, Proceedings of the 1994 Second CAD-Based Vision Workshop, pp. 220-227, February 1994;
[13] Curless, B., Levoy, M., A Volumetric Method for Building Complex Models from Range Images, Proceedings of SIGGRAPH '96;
[14] C. Bregler, Video Based Animation Techniques for Human Motion, SIGGRAPH '00 Course 39: Image-based Modeling and Rendering; and
[15] Paul F. Debevec, Camillo J. Taylor, Jitendra Malik, Modeling and Rendering Architecture from Photographs: A Hybrid Geometry-and Image-based Approach, Proceedings of SIGGRAPH '96.
The common problem with image-based methods is occurrence of holes in the resulting image. Unlike polygonal models that are ‘continuous’ in the sense that the object surface is linearly interpolated into the interior of all the polygons (normally, triangles), image-based and point-based representations provide ‘discrete’ approximations of the object. In case of image-based representations, the object surface is, in fact, approximated with small colored squares, i.e. shifted pixels of reference images. When viewing direction differs substantially from the normal direction to each of the reference image planes, projections of the approximating squares generally do not completely cover the projection of the object surface. Let as call such holes the holes of the first type. Another source of holes in the resulting image for image-based representations is the fact that some parts of the surface may be not visible in all of the reference images, but become visible for some viewpoints (holes of the second type). These holes are due to insufficient information contained in a particular image-based representation.
Relief texture method [1] suppresses holes of the first type by using an analog of linear interpolation, which may lead to distortions and artifacts, since interpolation is performed in the two-dimensional (2D) projection of the object rather than in 3D space. More importantly, holes of the second type can only be treated the same way under this approach. Since the method of [1] uses only 6 reference images, that is, projections of the object on the circumscribing cube faces, this imposes serious restrictions on this method application to complex shapes when there exist points invisible from all six cube faces. This approach was chosen to maximize rendering speed, namely by using fast prewarping (geometric transformation equivalent to the change of viewing direction under the orthographic projection), but it leads to quality degradation.
Layered depth images (LDI) [2] are data structure designed to avoid the problem with holes of the second type. LDI is an image whose pixels contain all the object points projecting to a fixed location in the reference image plane. Fast prewarping algorithm of [1] applies here as well. However, problems with holes of the first type remain. Splatting (first introduced in [10]) is used to solve the problem of holes of the first t
Han Mahn-jin
Zhirkov Alexander Olegovich
Pappas Peter-Anthony
Zimmerman Mark
LandOfFree
Hierarchical image-based representation of still and... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Hierarchical image-based representation of still and..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Hierarchical image-based representation of still and... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3343355