Tensor Oriented No-Reference Light Field Image Quality Assessment.

Area of Science:

• Computer Vision
• Image Processing
• Multimedia Systems

Background:

• Light field image (LFI) quality assessment is crucial for immersive media.
• LFIs present unique multi-dimensional challenges due to spatial and angular characteristics.
• Existing quality assessment methods struggle with the complexity of LFIs.

Purpose of the Study:

• To propose a novel no-reference light field image quality evaluator (Tensor-NLFQ).
• To address the multi-dimensional quality assessment problem in LFIs.
• To improve the accuracy and robustness of LFI quality evaluation.

Main Methods:

• Utilizing tensor theory to model LFIs as low-rank 4D tensors.
• Applying Tucker decomposition to extract principal components from sub-aperture view stacks.
• Developing Principal Component Spatial Characteristic (PCSC) for spatial quality.
• Introducing Tensor Angular Variation Index (TAVI) for angular consistency.

Main Results:

• The proposed Tensor-NLFQ effectively measures spatial quality using PCSC.
• TAVI accurately quantifies angular consistency in LFIs.
• Tensor-NLFQ demonstrates superior performance over state-of-the-art methods on multiple databases.

Conclusions:

• Tensor-NLFQ provides a robust and accurate solution for no-reference LFI quality assessment.
• The tensor-based approach effectively handles the multi-dimensional nature of LFIs.
• This method advances the field of immersive media quality evaluation.