Lightweight Web3D modeling by finding and reusing repeated components

Abstract

This paper presents a new method that can largely compress massive models which consist of a wide range of connected components. Its effectiveness relies mainly on the number and the complexity of repeated components being found in the input model. Comparing with the state-of-the-art algorithm of 3D model compression based on the reuse of repeated components, our method can find more repeated components, both efficiently and precisely, so that high compression ratio is achieved with no further compression of the unique components and transformation matrices. By employing reflection-invariant transformation and other optimization means during the alignment preprocessing of 3D models, we solve some problems existing in previous methods simply, like PCA ambiguities. Especially thanks to the matching scheme based on voxelization, our method itself is robust when confronting the situation that covariance matrix of PCA is degenerated. Experimental results show that our method reduces considerably and stably the number of connected components in 3D models than the state-ofthe-art algorithm so as to higher compression efficiency, and saves time around 20 times on average.