COBRA: Compression of basis of PCA represented animations

Abstract

In this paper we present an extension of dynamic mesh compression techniques based on PCA. Such representation allows very compact representation of moving 3d surfaces, however it requires some side information to be transmitted along with the main data. The biggest part of the side information is the PCA basis, and since the data can be encoded very efficiently, the size of the basis cannot be neglected when considering the overall performance of a compression algorithm. We present a pioneering work in this area, as none of the papers about PCA based compression really addresses this issue. We will show that for an efficient and accurate encoding there are better choices than even sophisticated algorithms such as LPC. We will present results showing that our approach can reduce the size of the basis by 90% with respect to direct encoding, which can lead to a 25% increase of performance of the compression algorithm without any significant loss of accuracy.