| Abstract: |
Texture mapping is an effective tool commonly used in 3D graphic
applications, especially when desired surface details are expensive to
achieve by solely using geometry. This talk addresses the problem of
efficiently streaming/storing textured 3D models in a resource
constrained
environment. For transmission/storage purposes, both the mesh and the
texture are progressively compressed to generate multi-resolution
representations. It is noticed that the traditional distortion metric,
the
peak signal-to-noise ratio (PSNR), fails in measuring the quality
degradation of the compressed textured models. Hence we propose a novel
and
fast quality measure (FQM) to properly capture the visual quality of the
textured model. The computation of FQM does not require extra
processing,
making it attractive for real-time applications. Based on FQM, we
present a
bit-allocation algorithm that distributes the source bits between the
mesh
and the texture to maximize the visual quality of the model.
|