Recently, the research team led by Professor Cao Weifeng from the College of Electrical and Information Engineering at Zhengzhou University of Light Industry (ZZULI) published a research paper titled "DACESR: Degradation-Aware Conditional Embedding for Real-World Image Super-Resolution" in IEEE Transactions on Image Processing, a top international journal in the field of image processing and analysis.

Against the backdrop of the rapid development of computer vision and multimodal AI, image super-resolution reconstruction, as a core research direction in image processing, is widely applied in numerous real-world scenarios such as surveillance and control systems, medical imaging, remote sensing and mapping, and high-definition image restoration. In recent years, multimodal large models have shown excellent abilities in addressing image super-resolution tasks by leveraging text as conditional information, yet their performance on degraded images remains limited. Firstly, this paper reappraises the performance of the Recognize Anything Model (RAM) on degraded images by calculating text similarity, and finds that directly using contrastive learning to fine-tune RAM in degraded space fails to achieve acceptable results. To address this issue, this paper introduces a degradation selection strategy and proposes a Real Embedding Extractor (REE), significantly improving the recognition performance on degraded image content via contrastive learning. Furthermore, via the Conditional Feature Modulator (CFM), the high-level semantic information extracted by REE is integrated into a powerful Mamba-based network. This enables the network to effectively leverage pixel-level information to restore image textures and produce visually pleasing reconstruction results. Extensive experiments demonstrate that the REE can effectively help image super-resolution networks to achieve a balance between fidelity and perceptual quality, while highlightingthe great potential of Mamba in addressing real-world image super-resolution tasks.

ZZULI is the first affiliated institution.Graduate student Lei Xiaoyan is the first author, and Cao Weifeng, Luo Biao and Liang Hui are the corresponding authors.

IEEE Transactions on Image Processing (IEEE TIP) is a top international journal of image processing and analysis, published by the Institute of Electrical and Electronics Engineers (IEEE). The journal mainly features cutting-edge research results in the fields of image, video and multidimensional signal processing, covering various areas such as mathematical modeling, image enhancement and restoration, coding transmission, biomedical imaging, and remote sensing. It is categorized as a CAS Q1 Top journal, with a 2025 impact factor of 13.7.

Journal article link: https://ieeexplore.ieee.org/abstract/document/11433537

Citation format: X. Lei, W. Zhang, B. Luo, H. Liang, W. Cao and Q. Lin, "DACESR: Degradation-Aware Conditional Embedding for Real-World Image Super-Resolution," in IEEE Transactions on Image Processing, vol. 35, pp. 2997-3008, 2026, doi: 10.1109/TIP.2026.3671639.