GenAI-Driven Unsupervised Denoising for Consumer Device Imagery

In the era of ubiquitous consumer electronics, image transmission across various communication channels has become an indispensable part of life. However, noise interference during transmission affects image quality, consumer satisfaction, and device functionality. To address this issue, we propose an unsupervised image denoising framework based on generative adversarial networks, which is applicable to consumer electronic devices under various communication channels and improves the quality of images received by users. We introduce a multilevel feature interaction model that utilizes a cross-attention mechanism to efficiently integrate multiscale feature information, preserve shallow details, and incorporate deep semantics. Extensive experiments show that our proposed framework exhibits excellent performance in a variety of realistic noisy scenarios, effectively removing a wide range of noise intensities while preserving image details and structural information, thus significantly improving the visual quality of content on consumer devices.