Tensor network methods for quantum-inspired image processing and classical optics

Tensor network methods strike a middle ground between fully-fledged quantum computing and classical computing, as they take inspiration from quantum systems to significantly speed up certain classical operations. Their strength lies in their compressive power and the wide variety of efficient algorithms that operate within this compressed space. In this work, we focus on applying these methods to fundamental problems in image compression and processing and classical optics such as wave-front propagation and optical image formation, by using directly or indirectly parallels with quantum mechanics and computation. These quantum-inspired methods are expected to yield faster algorithms with applications ranging from astronomy and earth observation to microscopy and classical imaging more broadly.

💡 Research Summary

This paper presents a comprehensive exploration of applying tensor network (TN) methods, inspired by concepts from quantum mechanics and computation, to fundamental problems in image processing and classical optics. The core premise is that TNs occupy a productive middle ground between fully-fledged quantum and classical computing, leveraging quantum-inspired mathematical structures to achieve significant acceleration for specific classical operations, primarily through their exceptional compressive power and efficient algorithms operating within the compressed space.

The work begins by establishing a theoretical foundation, drawing parallels between natural images and physical systems. It highlights the scale-invariant, power-law statistics of natural images and frames iterative downsampling as an analogue to a real-space renormalization group (RG) flow. This perspective allows for analyzing images through the lens of statistical mechanics, considering entropy scaling and correlation structure in the large-system limit.

A major technical contribution is the detailed comparison of two primary methods for encoding images into quantum-inspired tensor objects: the Flexible Representation of Quantum Images (FRQI) and the Quantics (or quantized tensor train) representation. FRQI encodes an image into a multi-qubit state by separating pixel intensity and position information, often using space-filling curves like the Hilbert curve to preserve locality. The Quantics representation factorizes spatial indices into their binary components, intrinsically building scale separation into the tensor’s index structure. Both methods transform a 2D image matrix into a high-order tensor, which is then compressed using TN decompositions.

The authors rigorously evaluate the compression performance of Matrix Product States (MPS) and Tensor Trains (TT) applied to these encodings using a representative natural image. They analyze the scaling of the mean squared error (ϵ) with system size (L) and the inverse compression ratio (ξ). Their numerical findings reveal a finite-size scaling law: ϵ(L) = L^(-κ) φ(ξ). In the efficient compression regime (ξ « 1), φ(ξ) ~ ξ^(-α) with α ≈ 0.75 and κ ≈ 1.5. This implies that the reconstruction error becomes essentially independent of system size, while the achievable compression ratio at a fixed error grows polynomially with L (C ~ L²). This behavior starkly contrasts with local compression schemes like JPEG, which have a hard maximum compression limit, underscoring the fundamental advantage of TN-based global compression for data with multi-scale correlations.

The second half of the paper shifts focus to applying these TN representations to classical optics simulations, such as wave-front propagation and optical image formation. A key insight is drawing formal parallels between Fourier optics and quantum mechanics. The authors propose constructing efficient approximations of classical optical propagation operators (e.g., large-aperture diffraction) by finding local Hamiltonians whose time evolution represents these operations. This connection allows the manipulation of high-dimensional optical states within the compressed TN space, potentially leading to exponentially faster simulations compared to conventional methods like direct Fourier transforms on full grids. The vision is to use TNs not just as static compression tools but as dynamic, operational frameworks within an entire image processing pipeline, enabling accelerated simulation, filtering, and analysis.

In summary, the paper demonstrates that tensor network methods, fueled by ideas from quantum information science, offer a powerful and promising framework for advancing classical computing tasks in image science and computational optics. By bridging concepts from statistical physics, information theory, and quantum computation, it lays the groundwork for developing faster algorithms with broad applications in astronomy, Earth observation, microscopy, and general imaging.