Max360IQ: Blind omnidirectional image quality assessment with multi-axis attention

2025-02-01

Pattern Recognition

AI Summary

Non-uniform distortion modeling and severe equi-rectangular projection (ECP) geometric distortion hinder blind no-reference quality assessment of 360° panoramic images. Method: We propose the first end-to-end blind 360° IQA model, incorporating a multi-axis attention mechanism to jointly encode spherical geometry, cross-latitudinal semantic dependencies, and local-global distortion responses; integrating spherical coordinate embedding and hierarchical distortion-sensitive feature extraction; and enabling the first adaptive perception of ECP-induced distortions—without requiring distortion priors or full-reference images. Contribution/Results: The model performs direct regression of subjective quality scores. Evaluated on OVQ and SIAT-360IQ benchmarks, it achieves state-of-the-art performance with PLCC ≥ 0.92, significantly outperforming both leading 2D IQA methods and existing 360° IQA approaches.

Technology Category

• Computer Vision: 3D Computer Vision
• Machine Learning: Calibration & Uncertainty Quantification
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• Security and Privacy: Large-scale security measurements

Problem

• Assessing quality of omnidirectional images
• Handling non-uniform image distortions
• Improving user experience with 360-degree images

Innovation

• Multi-axis attention for spatial interaction
• Multi-scale feature integration module
• Deep semantic guidance for quality prediction