MFAN: A Multi-Projection Fusion Attention Network for No-Reference and Full-Reference Panoramic Image Quality Assessment
Huanyang Li, Xinfeng Zhang · IEEE Signal Processing Letters · 2023
Panoramic images can provide viewers with a 360° perspective of the scene, and create an immersive experience to viewers, which makes it plays an important role in virtual reality (VR) applications. However, different from traditional 2D images, panoramic images are projected into 2D plane to process, e.g., compression, and viewed in VR devices, which makes their quality assessment be more challenging. In this paper, we propose a Multi-projection Fusion Attention Network (MFAN) to improve the accuracy of panoramic image quality assessment (PIQA). In particular, we propose to extract features from 2D plane images generated from multiple projection methods to overcome the distortions caused by a single projection. Furthermore, we design an attention module to improve the efficiency of features in PIQA task. In addition, our model can be utilized in both full refer-ence and no reference scenarios. Extensive experimental results show that our proposed MFAN can significantly outperform the existing IQA methods.