Lightweight Width-Depth Scalable Implicit Neural Representation for Progressive Image Compression

Qingyu Mao, Wenming Wang, Yongsheng Liang, Chenhu Xiao, Fanyang Meng, Gwanggil Jeon

IEEE Transactions on Consumer Electronics (TCE)

paper

This paper proposes a lightweight width-depth scalable implicit neural representation (INR) architecture for progressive image compression, which enables flexible progressive image reconstruction by adjusting the model’s width and depth, and achieves an optimal balance between compression efficiency, computational lightweight and architectural scalability for image compression tasks.

No‐Reference Image Quality Assessment: Past, Present, and Future

Qingyu Mao, Shuai Liu, Qilei Li, Gwanggil Jeon, Hyunbum Kim, David Camacho

Expert Systems

paper

This review article offers a comprehensive and systematic overview of the no-reference image quality assessment (NR-IQA) field, tracing its developmental journey from traditional handcrafted feature-based methods to modern deep learning-driven approaches, thoroughly analyzing the core current challenges including poor generalization across diverse distortions, limited adaptability to real-world complex scenes and content variations, and further outlining promising future research directions such as multimodal fusion, lightweight model design, cross-domain adaptation and self-supervised learning for more robust and practical NR-IQA systems.

CT and MRI image fusion via coupled feature-learning GAN

Qingyu Mao, Wenzhe Zhai, Xiang Lei, Zenghui Wang, Yongsheng Liang

Electronics

paper

This paper proposes a novel coupled feature-learning generative adversarial network (GAN) for CT and MRI image fusion, which is specially designed to extract and adaptively fuse the complementary anatomical structural features of CT images and the detailed soft tissue texture features of MRI images via its coupled feature-learning mechanism, effectively alleviates the problems of feature loss and information distortion in traditional medical image fusion methods, generates high-quality fused images integrating the advantages of both modal images, and verifies through extensive experiments that the proposed model outperforms the state-of-the-art image fusion approaches, showing promising application potential for clinical medical diagnosis.