KIAM Main page Web Library  •  Publication Searh   

Conference material: "Proceedings of the International Conference on Computer Graphics and Vision Graphicon (19-21 September 2023, Moscow)"
Authors: Matveev S.A., Kurilovich A.A.
Utilization of Tensor Decompositions for Video-compression
In this work, we provide a study of video compression with the use of tensor train and Tucker decompositions. We measure the quality of compression with classical PSNR and SSIM metrics. Our approach allows us to control the quality of compressed video through the analytical evaluation of tensor decomposition ranks using the target value of PSNR. We achieve this aim because the PSNR is naturally related to the value of relative error in the Frobenius norm, which can be controlled for both tensor train and Tucker decompositions. In case of tensor train decomposition, we evaluate the idea of adding additional virtual dimensions and show that this trick allows us to improve the quality of compression without adding non- negligible additional errors. We discuss the advantages and visible artifacts introduced by the tensor-based algorithms to video compression and compare our results with industrial standards.
video compression, tensor decomposition, tensor train, Tucker decomposition
Publication language: english,  pages: 8 (p. 582-589)
English source text:
Export link to publication in format:   RIS    BibTeX
About authors:
  • Matveev Sergey Alexandrovich, Moscow State University; Institute of Numerical Mathematics RAS
  • Kurilovich Aleksandr A.,  Skolkovo Institute of Science and Technology