WSEAS Transactions on Signal Processing
Print ISSN: 1790-5052, E-ISSN: 2224-3488
Volume 15, 2019
Complexity Evaluation of Tensor Decomposition through 3D Inverse Spectrum Pyramid in respect of Deterministic Orthogonal Transforms
Authors: , ,
Abstract: Recently, orthogonal 3D tensor decompositions are widely involved in the processing of various kinds of 3D data such as multimedia signals, correlated image sequences, etc. The methods for tensor decomposition could be divided into two main groups: statistical, based on various modifications of the Principal Component Analysis and the Singular Value Decomposition, and deterministic, based on the pyramidal 3D Discrete Wavelet Transform decompositions, Curvelet/Contourlet Discrete Transform and the Shearlet Discrete Transform. The methods from the first group surpass these from the second in the higher decorrelation of the decomposition components, but these from the second group have much lower computational complexity. In this work are compared the structures and is evaluated the computational complexity of the tensor decompositions, mentioned above, with the decomposition developed by the authors, which is based on the Reduced 3D Inverse Spectrum Pyramid (3D-RISP). The comparison results show that: the global computational complexity of the 3D-RISP is much lower than that of the remaining pyramidal decompositions; the compared schemes for recursive calculation have similar structures but this of the RISP does not need operations of the kind decimation and interpolation which are the reason for distortions in the reconstructed tensors.
Search Articles
Keywords: Deterministic pyramidal decompositions, Computational complexity, Recursive calculation, 3D Inverse Spectrum Pyramid, 3D Walsh-Hadamard transform
Pages: 142-148
WSEAS Transactions on Signal Processing, ISSN / E-ISSN: 1790-5052 / 2224-3488, Volume 15, 2019, Art. #17