Cover imageMultiresolution Representation and Reconstruction of Adaptively Sampled Images


Lucia Darsa and Bruno Costa



We present a method for sampling, structuring and reconstructing progressively refined images effectively. This has direct applications to progressive refinement of the result of slow renderers, such as ray tracers, and to the progressive transmission of images on slow media, as in world wide web browsers. Sampling images adaptively with respect to frequency at the back end provides a more compact representation. These samples are then structured at the front end using Voronoi or Delaunay tesselations, and reconstructed with C0 continuity taking advantage of standard 3D acceleration hardware. The structuring scheme is extended to handle multiresolution images, with optional antialiasing, allowing the representation of an arbitrary level of detail without penalizing the reconstruction.

L. Darsa and Bruno Costa (1996). Multiresolution Representation and Reconstruction of Adaptively Sampled Images. In SIBGRAPI '96 Proceedings, pages 321-328. Zipped Postscript file (800Kb)

Some of the color images of the paper are also here:

Figure 6

Figure 7

Figure 10

 Adding Samples

dice film stripClick on the thumbnail to view a QuickTime movie showing the adaptive addition of samples to an image. Each frame contains 20 samples more than the previous, and the reconstruction of the image is based on Gouraud shading. Note that the movie contains artifacts due to the QuickTime lossy spatial and temporal compression (2,027Kb).

Some of the frames of the movie are shown in JPEG format below. Click on the image to see them in full resolution. If you want to see the images without any kind of lossy compression, click on the frame name to get them in TIFF format.

very low res dice

low res dice

medium res dice


(20 samples)

frame 70

(1400 samples)

frame 140

(2800 samples)


 e-mail: or

www: or


The background of the pages is an actual cyclic Voronoi diagram, generated by the software the authors developed for incremental Voronoi tesselation (followed by a little post-processing :-).The center of each Voronoi cell is visible if you look carefully. The software is available from the authors upon request.

  Last updated on August 29, 1996