Abstract:A flexible transform-based tensor product named $\star_{\rm{QT}}$-product for $L$th-order ($L\geq 3$) quaternion tensors is proposed. Based on the $\star_{\rm{QT}}$-product, we define the corresponding singular value decomposition named TQt-SVD and the rank named TQt-rank of the $L$th-order ($L\geq 3$) quaternion tensor. Furthermore, with orthogonal quaternion transformations, the TQt-SVD can provide the best TQt-rank-$s$ approximation of any $L$th-order ($L\geq 3$) quaternion tensor. In the experiments, we have verified the effectiveness of the proposed TQt-SVD in the application of the best TQt-rank-$s$ approximation for color videos represented by third-order quaternion tensors.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the limitations of the existing quaternion tensor singular value decomposition (Qt - SVD) methods. Specifically:
1. **Limited scope of application**: The existing Qt - SVD methods are only applicable to third - order quaternion tensors and cannot handle higher - order quaternion tensors.
2. **Fixed transformation method**: The transformations (such as the discrete Fourier transform, DFT) on which Qt - SVD depends are fixed, which may lead to insufficient flexibility in application and inability to adapt to different types of color video data.
To solve these problems, the paper proposes a quaternion tensor singular value decomposition based on flexible transformation (TQt - SVD). The main contributions of this method are as follows:
- **Extension to high - order tensors**: TQt - SVD can be applied to any L - order (L ≥ 3) quaternion tensors, not just limited to third - order tensors.
- **Flexible transformation selection**: TQt - SVD allows the use of any invertible quaternion transformation, rather than being fixed to a specific transformation, thereby improving the flexibility and adaptability of the method.
Through these improvements, TQt - SVD can provide more effective low - rank approximation in the optimal TQt - rank - s approximation of color videos, and its superiority has been verified in experiments, especially when using data - driven transformations.
### Summary of mathematical formulas
- **Quaternion facewise product (⋆QF - product)**:
\[
\dot{F}=\dot{A} \star_{\text{QF}} \dot{B} \in \mathbb{H}^{N_1 \times N_2 \times \cdots \times N_L}
\]
Satisfying that each frontal slice satisfies:
\[
\dot{F}(:,:,n_3,\ldots,n_L)=\dot{A}(:,:,n_3,\ldots,n_L) \cdot \dot{B}(:,:,n_3,\ldots,n_L)
\]
- **Quaternion tensor product (⋆QT - product)**:
\[
\dot{T}=\dot{A} \star_{\text{QT}} \dot{B}=(\hat{\dot{A}} \star_{\text{QF}} \hat{\dot{B}}) \times_3 \dot{T}_3^{-1} \times \cdots \times_L \dot{T}_L^{-1}
\]
where:
\[
\hat{\dot{A}}=\dot{A} \times_3 \dot{T}_3 \times \cdots \times_L \dot{T}_L
\]
\[
\hat{\dot{B}}=\dot{B} \times_3 \dot{T}_3 \times \cdots \times_L \dot{T}_L
\]
- **TQt - SVD decomposition**:
\[
\dot{A}=\dot{U} \star_{\text{QT}} \dot{D} \star_{\text{QT}} \dot{V}^H
\]
where \(\dot{U}\) and \(\dot{V}\) are unitary quaternion tensors, and \(\dot{D}\) is an f - diagonal quaternion tensor.
- **TQt - rank definition**:
\[
\text{rank}_{\text{TQt}}(\dot{A})=\left\{k \mid \| \dot{D}(k,k,:,\ldots,:) \|_F > 0\right\}, \quad k \in [K], \quad K = \min(N_1, N_2)
\]