Abstract
Signal processing over hypercomplex numbers arises in many optical imaging applications. In particular, spectral image or color stereo data are often processed using octonion algebra. Recently, the eight-band multispectral image phase recovery has gained salience, wherein it is desired to recover the eight bands from the phaseless measurements. In this paper, we tackle this hitherto unaddressed hypercomplex variant of the popular phase retrieval (PR) problem. We propose octonion Wirtinger flow (OWF) to recover an octonion signal from its intensity-only observation. However, contrary to the complex-valued Wirtinger flow, the non-associative nature of octonion algebra and the consequent lack of octonion derivatives make the extension to OWF non-trivial. We resolve this using the pseudo-real-matrix representation of octonion to perform the derivatives in each OWF update. We demonstrate that our approach recovers the octonion signal up to a right-octonion phase factor. Numerical experiments validate OWF-based PR with high accuracy under both noiseless and noisy measurements.
Introduction
Phase retrieval (PR) is a fundamental challenge in computational imaging, where the objective is to reconstruct a signal from the magnitude of its linear measurements. While extensively studied for complex-valued signals, many modern sensing modalities, such as multispectral imaging, involve multi-channel data. Processing these channels independently often fails to exploit the inherent inter-channel correlations that exist in real-world signals.
Octonions, an 8-dimensional normed division algebra, provide a natural framework for representing 8-channel signals. However, octonions are non-associative, meaning $(a \cdot b) \cdot c \neq a \cdot (b \cdot c)$ in general. This property breaks standard calculus and optimization techniques, such as the chain rule, making the development of gradient-based algorithms like Wirtinger Flow (WF) for octonions a significant mathematical challenge.
Methodology
To address the non-associativity of octonions, we propose Octonion Wirtinger Flow (OWF). The core innovation lies in the pseudo-real-matrix representation. By mapping octonion variables into $8 \times 8$ real-valued matrices, we can perform optimization in a space where standard real-valued derivatives are well-defined, while still preserving the underlying octonion algebraic structure.
The OWF algorithm follows a two-stage approach:
- Spectral Initialization: We adapt the power method to the octonion domain to generate a high-quality initial estimate, which is crucial for avoiding local minima in non-convex PR problems.
- Iterative Refinement: We minimize a squared loss function between observed and estimated intensities. We prove that the signal can be recovered up to a global right-octonion phase factor, which represents the fundamental ambiguity of the octonion PR problem.
Experimental Results
We evaluated OWF using both synthetic data and real-world multispectral images from the CAVE dataset.
Sample Efficiency and Convergence
In synthetic experiments with Gaussian signals, OWF demonstrated superior sample efficiency compared to traditional methods. As shown in Figure 1, OWF achieves a success rate of 1 (perfect recovery) at a sampling ratio of $m/n \approx 10$. In contrast, a "Concatenated WF" approach—which treats the 8 channels as a single long complex vector—requires a significantly higher sampling ratio ($m/n > 20$) to achieve similar performance. Furthermore, OWF exhibits linear convergence, with the distance to the true signal decreasing exponentially over iterations.
Multispectral Image Recovery
We applied OWF to a $32 \times 32$ crop of the CAVE dataset, selecting 8 spectral bands. OWF achieved a Peak Signal-to-Noise Ratio (PSNR) of 39.01 dB, visually and numerically outperforming a standard real-valued Gradient Descent (GD) approach, which only reached 24.16 dB. Figure 2 illustrates the high fidelity of the OWF reconstruction across the spectral bands, capturing fine details that are lost in standard GD recovery.
Conclusion
This work introduces the first framework for Octonion Phase Retrieval. By leveraging the pseudo-real-matrix representation, we overcome the non-associativity of octonions to develop the OWF algorithm. Our results show that joint processing in the octonion domain significantly improves sample efficiency and reconstruction quality for multi-channel signals. Future work will explore the theoretical phase transition boundaries and the extension of these methods to even higher-dimensional algebras like sedenions.
Summary updated