Cubemap-Based Perception-Driven Blind Quality Assessment for 360-degree Images.

Area of Science:

• Computer Vision
• Image Processing
• Human-Computer Interaction

Background:

• 360-degree image quality assessment (360-IQA) faces challenges due to format-specific distortions, particularly in equirectangular projection (ERP) images.
• The performance imbalance between ERP and other 360-degree image formats hinders applications like compression and assessment.

Purpose of the Study:

• To propose a novel blind 360-IQA framework that addresses the performance imbalance caused by ERP image distortions.
• To enhance the accuracy and effectiveness of 360-IQA by incorporating human visual attention mechanisms.

Main Methods:

• Utilized cubemap projection (CMP) with six faces for omnidirectional viewing of 360-degree images.
• Established a multi-distortion visual attention quality dataset for 360-degree images as a benchmark.
• Developed a perception-driven blind 360-IQA framework considering human attention behavior and extracting quality feature subsets.

Main Results:

• The proposed CMP-based blind 360-IQA framework demonstrated superior performance compared to state-of-the-art methods.
• Cross-dataset validation confirmed the framework's effectiveness and robustness.
• The framework shows potential for integration with new feature extraction methods to further boost 360-IQA performance.

Conclusions:

• The novel blind 360-IQA framework effectively handles ERP image distortions using CMP.
• Incorporating human attention significantly improves the accuracy of 360-IQA.
• The proposed framework offers a promising solution for future 360-degree image quality assessment applications.