Investigating the Rate-Distortion Performance of a Wavelet-Based Mesh Compression Algorithm by Perceptual and Geometric Distortion Metrics

The rate-distortion performance of wavelet-based mesh compression algorithms is usually only evaluated in a purely geometric sense, by measures such as the Root Mean Square Error and the Hausdorff distance, which do not capture the human visual perception of distortion. This lack of quantitative information about the perceptual effects of wavelet compression has prompted us to present a more complete evaluation of a classic waveletbased mesh compression algorithm, by measuring its rate-distortion performance both with geometric metrics (the Hausdorff distance and Root Mean Square Error) and perceptual metrics (the recently introduced Mesh Structural Distortion Measure 2 (MSDM2) and the Mean Opinion Scores (MOS) that we obtained by conducting a subjective experiment with human observers), where the rate is measured as the percentage of wavelet coefficients used in reconstruction. The MSDM2 has already been proven to outperform other existing perceptual metrics for several different types of distortions, but this is the first time that it has been tested for this type of geometric distortion in a real-use case scenario. We found that, in this context, the MSDM2 generally correlates well with the MOS but seems to under-estimate the perceptual error in cases of low-frequency (large scale) shape distortion. Due to the disparities in the distortion values produced by the tested distortion metrics, we also conclude that a complete evaluation of any mesh compression algorithm should include several different distortion metrics, to allow the developers and users of these compression algorithms to make more informed decisions about the applicability of those algorithms in different application areas.