Quantum Light Classifies Images Without Needing a Conventional Camera

Giorgio Minati and colleagues at the Sapienza University of Rome have created a camera-free quantum-optical image classifier that performs inference using Hong-Ou-Mandel interference of single photons. The new approach replaces traditional pixel-by-pixel acquisition with a single measurement, realising both a single-perceptron neuron and a two-neuron shallow network with high accuracy and strong resistance to noise. Performance remains consistent regardless of input resolution, a feat unattainable with classical systems, and suggests potential for advancements in areas such as remote object recognition and low-signal sensing. The system provides a flexible set of tools for machine learning applications that overcome the limitations of classical computing.

Quantum interference enables pixel-independent image classification with high accuracy

Accuracy on benchmark datasets reached 98.1%, a figure previously unattainable without scaling performance to input resolution. This camera-free quantum-optical image classifier achieves performance independent of the number of pixels, a characteristic impossible in conventional systems reliant on pixel-by-pixel acquisition. The demonstrated system utilises Hong-Ou-Mandel (HOM) interference, a quantum phenomenon where two photons ‘interfere’ with each other, to directly compare an image with learned templates, replacing complex digital processing with a single measurement.

Realising both a single-perceptron neuron and a two-neuron shallow network, this work establishes a pathway towards neuromorphic photonic processors for applications including low-light imaging and remote sensing. Researchers further validated the system’s resilience by demonstrating consistent performance even with substantial experimental noise; the classifier maintained high accuracy despite real-world imperfections in the optical setup. Beyond single images, a two-neuron shallow network was successfully implemented, indicating the potential for scaling up the architecture to tackle more complex classification tasks; this network processed information with the same pixel-independent accuracy.

Importantly, the system achieved this 98.1% accuracy utilising a fixed measurement budget, meaning the number of photons required for analysis remained constant regardless of image size. This contrasts sharply with conventional systems where larger, higher-resolution images demand exponentially more processing power and photons. However, these figures currently represent performance on controlled benchmark datasets, and significant engineering challenges remain before this technology can be deployed in dynamic, real-world scenarios with unpredictable lighting or complex backgrounds.

Quantum optical neurons demonstrate robust pattern recognition with minimal resources

Scientists are realising a single-perceptron quantum optical neuron and a two-neuron shallow network, achieving high accuracy on benchmark datasets with strong robustness to experimental noise and minimal hardware complexity. With a fixed measurement budget, performance remains insensitive to input resolution, demonstrating intrinsic robustness to the number of pixels, which would be impossible in a classical framework. This approach paves the way toward neuromorphic quantum photonic processors capable of extracting task-relevant information directly from Hong-Ou-Mandel interference, with promising applications in remote object recognition, low-signal sensing, and photon-starved biological microscopy.

The past decade has witnessed unprecedented advancements in Machine Learning, yielding outstanding results across different domains ranging from natural language processing to photon-starved biological microscopy. Optical and quantum technologies offer routes for high-dimensional, parallel information processing suited for imaging tasks, addressing limitations of classical hardware. Quantum photonic platforms compute inner products and offer potential for energy-efficient inference.

A camera-free quantum-optical image classifier performs inference at the measurement layer using Hong, Ou, Mandel (HOM) interference of spatially programmable single photons. Two-photon coincidences report the overlap between an input image mode and a learned template, replacing pixel-resolved acquisition with a single global measurement.

Both a single-perceptron quantum optical neuron and a two-neuron shallow network were realised, achieving high accuracy on benchmark datasets with minimal hardware complexity. Performance remains insensitive to input resolution with a fixed measurement budget, demonstrating robustness to the number of pixels, an outcome impossible classically. This approach enables neuromorphic quantum photonic processors capable of extracting task-relevant information directly from HOM interference, with applications in remote object recognition, low-signal sensing, and photon-starved biological microscopy.

The rapid growth of machine learning is increasingly constrained by the energy and bandwidth limits of classical hardware. Optical and quantum technologies offer an alternative route, enabling high-dimensional, parallel information processing directly in the physical layer, particularly suited for imaging tasks. In this context, quantum photonic platforms provide both a natural mechanism for computing inner products and a promising path to energy-efficient inference in photon-limited regimes.

A camera-free quantum-optical image classifier performs inference directly at the measurement layer using Hong, Ou, Mandel (HOM) interference of spatially programmable single photons. Two-photon coincidences directly report the overlap between an input image mode and a learned template, replacing pixel-resolved acquisition with a single global measurement. A single-perceptron quantum optical neuron and a two-neuron shallow network were realised, achieving high accuracy on benchmark datasets with robustness to experimental noise and minimal hardware complexity. With a fixed measurement budget, performance remains insensitive to input resolution, demonstrating robustness to the number of pixels. This approach paves the way to efficient image processing across numerous applications, ranging from remote object recognition to photon-starved biological microscopy, in which the information to be manipulated is natively encoded in an optical field.

Researchers review how to use the Hong-Ou-Mandel effect to implement artificial neurons and shallow neural networks on an optical device. They consider two photons separately fed into a balanced beam splitter, encoding the input data X and the model weights λ in the spectral amplitudes of the photon spatial modes, through a pure state and a density operator ρ, respectively. The detection is performed by two bucket detectors with no spatial resolution, and by repeating the measurement, one obtains z(X) from the coincidence rate.

Optical and quantum technologies offer routes for high-dimensional, parallel information processing suited for imaging tasks, addressing limitations of classical hardware. Quantum photonic platforms compute inner products and offer potential for energy-efficient inference. A camera-free quantum-optical image classifier performs inference at the measurement layer using Hong, Ou, Mandel (HOM) interference of spatially programmable single photons. Two-photon coincidences report the overlap between an input image mode and a learned template, replacing pixel-resolved acquisition with a single global measurement.

Both a single-perceptron quantum optical neuron and a two-neuron shallow network were realised, achieving high accuracy on benchmark datasets with minimal hardware complexity. Performance remains insensitive to input resolution with a fixed measurement budget, demonstrating robustness to the number of pixels. This approach enables neuromorphic quantum photonic processors capable of extracting task-relevant information directly from HOM interference, with applications in remote object recognition, low-signal sensing, and photon-starved biological microscopy.

The rapid growth of machine learning is increasingly constrained by the energy and bandwidth limits of classical hardware. Optical and quantum technologies offer an alternative route, enabling high-dimensional, parallel information processing directly in the physical layer, particularly suited for imaging tasks. In this context, quantum photonic platforms provide both a natural mechanism for computing inner products and a promising path to energy-efficient inference in photon-limited regimes.

A camera-free quantum-optical image classifier performs inference directly at the measurement layer using Hong, Ou, Mandel (HOM) interference of spatially programmable single photons. Two-photon coincidences directly report the overlap between an input image mode and a learned template, replacing pixel-resolved acquisition with a single global measurement. A single-perceptron quantum optical neuron and a two-neuron shallow network were realised, achieving high accuracy on benchmark datasets with robustness to experimental noise and minimal hardware complexity. With a fixed measurement budget, performance remains insensitive to input resolution, demonstrating robustness to the number of pixels.

Researchers demonstrated a camera-free quantum-optical image classifier that uses Hong-Ou-Mandel interference with single photons to perform image recognition. This system performs inference directly at the measurement layer, replacing traditional pixel-by-pixel acquisition with a single measurement and achieving high accuracy on benchmark datasets. Importantly, performance remained consistent regardless of input resolution, indicating robustness to image size. The authors realised both a single-perceptron neuron and a two-neuron network, paving the way for neuromorphic quantum photonic processors.