Quantum machine learning is expected to be one of the potential applications that can be realized in the near future. Finding potential applications for it has become one of the hot topics in the quantum computing community. With the increase of digital image processing, researchers try to use quantum image processing instead of classical image processing to improve the ability of image processing. Inspired by previous studies on the adversarial quantum circuit learning, we introduce a quantum generative adversarial framework for loading and learning a quantum image. In this paper, we extend quantum generative adversarial networks to the quantum image processing field and show how to learning and loading an classical image using quantum circuits. By reducing quantum gates without gradient changes, we reduced the number of basic quantum building block from 15 to 13. Our framework effectively generates pure state subject to bit flip, bit phase flip, phase flip, and depolarizing channel noise. We numerically simulate the loading and learning of classical images on the MINST database and CIFAR-10 database. In the quantum image processing field, our framework can be used to learn a quantum image as a subroutine of other quantum circuits. Through numerical simulation, our method can still quickly converge under the influence of a variety of noises.

中文翻译：

---

用于在噪声环境中学习和加载量子图像的量子生成对抗网络

量子机器学习有望成为在不久的将来可以实现的潜在应用之一。为其寻找潜在应用已成为量子计算界的热门话题之一。随着数字图像处理的增多，研究人员尝试用量子图像处理代替经典图像处理来提高图像处理能力。受先前关于对抗性量子电路学习的研究的启发，我们介绍了一种用于加载和学习量子图像的量子生成对抗性框架。在本文中，我们将量子生成对抗网络扩展到量子图像处理领域，并展示了如何使用量子电路学习和加载经典图像。通过减少没有梯度变化的量子门，我们将基本量子构建块的数量从 15 个减少到 13 个。我们的框架有效地生成受比特翻转、比特相位翻转、相位翻转和去极化信道噪声影响的纯状态。我们在 MINST 数据库和 CIFAR-10 数据库上数值模拟经典图像的加载和学习。在量子图像处理领域，我们的框架可用于学习量子图像作为其他量子电路的子程序。通过数值模拟，我们的方法在各种噪声的影响下仍然可以快速收敛。我们的框架可用于学习量子图像作为其他量子电路的子程序。通过数值模拟，我们的方法在各种噪声的影响下仍然可以快速收敛。我们的框架可用于学习量子图像作为其他量子电路的子程序。通过数值模拟，我们的方法在各种噪声的影响下仍然可以快速收敛。