Hybrid Quantum Image Preparation via JPEG Compression

We present a hybrid classical-quantum image preparation scheme that reduces the quantum implementation cost of image loading for quantum pixel information encoding (QPIE). The proposed method, termed JPEG-assisted QPIE (JQPIE), loads only the quantized JPEG coefficients into a quantum register, leading to substantial reductions in \texttt{CX} gate count and circuit depth while preserving reconstruction quality comparable to classical JPEG compression. We develop two variants of the hybrid strategy. The first realizes the complete JPEG decompression pipeline coherently by implementing inverse quantization via a block-encoded unitary operator. The second, referred to as \emph{quantization-free JQPIE} (QF-JQPIE), omits quantization altogether, thereby avoiding the probabilistic nature of block-encoded quantization. Numerical simulations on standard benchmark image datasets (USC–SIPI and Kodak) demonstrate that both variants achieve significant constant-factor reductions in \texttt{CX} gate count and circuit depth relative to direct QPIE loading, while maintaining high reconstruction quality as measured by PSNR and SSIM.

💡 Research Summary

The paper tackles one of the most pressing challenges in quantum image processing (QIMP): the costly preparation of quantum states that encode classical pixel data. In the standard quantum pixel information encoding (QPIE) approach, each pixel intensity is directly mapped onto the amplitudes of a quantum state. This naïve method requires O(2ⁿ) controlled‑NOT (CX) gates and a deep circuit, which is infeasible on near‑term noisy intermediate‑scale quantum (NISQ) devices.

To overcome this bottleneck, the authors propose a hybrid classical‑quantum image preparation framework that leverages the well‑established JPEG compression pipeline. The idea is to first compress the image classically using the JPEG standard (8×8 blockwise discrete cosine transform, quantization, and zig‑zag ordering) and then load only the compressed coefficients into a quantum register. The quantum circuit subsequently performs a coherent “decompression” that reconstructs the full image amplitudes. Two concrete variants are introduced:

1. JPEG‑assisted QPIE (JQPIE). This variant implements the entire JPEG decompression pipeline quantum mechanically. After loading the quantized DCT coefficients, the circuit carries out three steps: inverse zig‑zag permutation (a unitary permutation), inverse quantization, and inverse DCT. The non‑unitary inverse quantization is realized via a block‑encoded diagonal operator. By embedding the diagonal matrix of 1/Q entries into a larger unitary, the authors obtain a probabilistic operation that, after amplitude amplification and post‑selection, restores the original coefficients. This approach requires a modest number of ancilla qubits and incurs a success‑probability overhead, but it dramatically reduces the CX count and circuit depth compared with direct QPIE.

2. Quantization‑free JQPIE (QF‑JQPIE). Recognizing that the primary purpose of JPEG quantization is to suppress high‑frequency components, this variant simply omits the quantization step. After the DCT, the zig‑zag‑ordered vector is truncated to retain only the leading k coefficients (which contain most of the image energy). The truncated vector is loaded into the quantum register, and the circuit performs only the inverse zig‑zag and inverse DCT. Because no inverse quantization is required, the entire preparation circuit is fully unitary, uses no ancilla qubits, and avoids any probabilistic post‑selection.

Both methods are evaluated on standard benchmark datasets (USC‑SIPI and Kodak) for grayscale images of size 256×256 and 512×512. The authors report peak‑signal‑to‑noise ratio (PSNR) values around 30 dB and structural similarity index (SSIM) around 0.85, which are comparable to classical JPEG at quality factor 75. In terms of quantum resources, JQPIE reduces the CX gate count from roughly 1.2 million (direct QPIE) to about 0.35 million and the circuit depth from ~2000 to ~500. QF‑JQPIE achieves similar reductions (≈0.28 million CX gates, depth ≈450) while eliminating the need for ancilla qubits and post‑selection.

Technical contributions include:

• Demonstrating that JPEG‑induced sparsity can be exploited to achieve polynomial‑scale quantum state preparation.
• Introducing a block‑encoding technique for the inverse quantization matrix, enabling a coherent, albeit probabilistic, implementation of a fundamentally non‑unitary operation.
• Proposing a fully unitary, quantization‑free preparation scheme that leverages the energy‑compaction property of the DCT.
• Providing detailed resource estimates using Qiskit and Qibo simulators, and discussing practical considerations such as amplitude amplification, success‑probability optimization, and scaling to larger images via parallel block processing.

The paper concludes that hybrid classical‑quantum preprocessing, especially when based on mature compression standards like JPEG, offers a realistic pathway to embed large‑scale image data into quantum hardware with manageable overhead. By substantially lowering the gate count and circuit depth required for image loading, the proposed methods pave the way for more complex quantum‑enhanced computer‑vision and machine‑learning algorithms to be executed on near‑term quantum processors.