Video dimension reduction and coding using Multiple Tensor Rank-R Decomposition

In this paper, we propose a dimension reduction algorithm for video coding based on tensor decomposition. Our Multiple Tensor Rank-R Decomposition (MTRD) algorithm uses rank-R tensors to repeatedly approximate the high dimensional video data, so that the dimension of the data can be greatly reduced. Through quantifying and coding these tensor decomposition coefficients, we can achieve a better video quality with the same compression ratio as H.264/AVC. In particular, we find that our method can implement a high-quality coding at a very low bit-rate for a class of videos with regular texture.