No-Reference Quality Assessment for 3D Synthesized Images Based on Visual-Entropy-Guided Multi-Layer Features...

Area of Science:

• Computer Vision
• Image Processing
• Human-Computer Interaction

Background:

• Multiview video plus depth (MVD) is key for free viewpoint video, using depth-image-based rendering (DIBR) for 3D scene synthesis.
• Inaccuracies in depth maps and DIBR lead to geometric distortions, degrading visual quality and user experience.
• Effective 3D synthesized image quality assessment (IQA) is crucial for evaluating content feasibility by simulating human visual perception.

Purpose of the Study:

• To propose a no-reference IQA metric for 3D synthesized images.
• To analyze multi-layer features guided by visual entropy for distortion assessment.
• To improve the accuracy of 3D synthesized image quality evaluation.

Main Methods:

• Developed a no-reference IQA metric utilizing visual-entropy-guided multi-layer feature analysis.
• Categorized geometric distortions into bottom-up (salient) and top-down (insignificant) visual attention layers based on energy entropy.
• Quantified salient distortion features using regional proportion and transition thresholds; analyzed insignificant distortion features via relative total variation and attention interactions.

Main Results:

• The proposed metric effectively integrates features from both bottom-up and top-down layers for comprehensive quality evaluation.
• Experimental results demonstrate superior performance compared to existing state-of-the-art methods.
• The metric accurately assesses the visual quality of 3D synthesized images.

Conclusions:

• The developed visual-entropy-guided multi-layer feature analysis metric provides a more perceptually relevant evaluation of 3D synthesized image quality.
• This approach offers a robust solution for assessing the quality of free viewpoint video content.
• The proposed IQA metric advances the field by offering a superior alternative for quality assessment in 3D imaging applications.