Rich Structural Index for Stereoscopic Image Quality Assessment

Area of Science:

• Computer Vision
• Image Processing
• Human-Computer Interaction

Background:

• Human visual system (HVS) perception of stereo images is influenced by viewing distance, crucial for stereoscopic image quality assessment (SIQA).
• Existing SIQA methods often neglect comprehensive human visual perception characteristics.
• There is a need for objective SIQA methods that better align with human visual perception.

Purpose of the Study:

• To propose a novel objective stereoscopic image quality assessment method.
• To enhance SIQA by integrating multi-scale human visual perception characteristics.
• To develop a Rich Structural Index (RSI) for improved SIQA accuracy.

Main Methods:

• Stereo image pairs processed through a Contrast Sensitivity Function (CSF) based image pyramid for multi-scale analysis.
• Local Luminance and Structural Index (LSI) extracted considering luminance and contrast masking.
• Singular Value Decomposition (SVD) used for Sharpness and Intrinsic Structural Index (SISI); gradient cross-mapping for Depth Texture Structural Index (DTSI).
• Support Vector Machine Regression based on Genetic Algorithm (GA-SVR) for final quality score prediction.

Main Results:

• The proposed Rich Structural Index (RSI) method demonstrates stable performance across four open databases.
• Experimental evaluations show a strong competitive advantage compared to state-of-the-art SIQA methods.
• The method effectively captures image distortions and perceptual characteristics.

Conclusions:

• The proposed RSI method provides a robust and effective approach for objective stereoscopic image quality assessment.
• Incorporating multi-scale perception characteristics and advanced signal processing techniques improves SIQA accuracy.
• The method offers a significant advancement in evaluating stereoscopic image quality aligned with human perception.