Accelerated MPEG Compression of Dynamic Polygonal Scenes

Dan S. Wallach
Sharma Kunapalli
Michael F. Cohen

This paper describes a methodology for using the matrix-vector multiply and scan conversion hardware present in many graphics workstations to rapidly approximate the optical flow in a scene. The optical flow is a 2-dimensional vector field describing the on-screen motion of each pixel. An application of the optical flow to MPEG compression is described which results in improved compression with minimal overhead.

Computer Graphics (Proc. SIGGRAPH 1994), July 1994

Compressed postscript version of the paper (greyscale figures) (180k)
PDF version of the paper (greyscale figures) (171k)

Color Figures (1056 x 720 pixels at full size)
Plate 1
1172k, 24-bit TIFF (exact original)
262k, 8-bit GIF (dithered)
74k, 24-bit JPEG (lossy)
Plate 2
601k, 24-bit TIFF (exact original)
90k, 8-bit GIF (dithered)
18k, 24-bit JPEG (lossy)

Postscript graphs showing more results than the paper, plus more info on the figures

Sample MPEG Video (653k)
a description of the video


Back up to papers index
Dan Wallach, CS Department, Princeton University