Virtual-view PSNR prediction based on a depth distortion tolerance model and support vector machine.

Area of Science:

• Computer Vision
• Video Processing
• Image Quality Assessment

Background:

• Virtual-view quality prediction is crucial for free viewpoint video (FVV) systems.
• Accurate prediction aids in optimizing depth video coding and virtual-view rendering.
• Existing methods may lack efficiency or accuracy in complex scenarios.

Purpose of the Study:

• To develop an efficient and accurate method for predicting virtual-view quality, specifically peak signal to noise ratio (PSNR).
• To analyze the impact of depth distortion on virtual-view quality and model this relationship.
• To compare the performance and computational complexity of the proposed method against existing approaches.

Main Methods:

• Analysis of depth distortion effects on virtual-view quality.
• Development of a depth distortion tolerance (DDT) model to define distortion ranges.
• Implementation of a support vector machine (SVM) model trained using the DDT model for quality prediction.

Main Results:

• The DDT model achieved an average Spearman's rank correlation coefficient of 0.8750 and root mean square error of 0.6137.
• The SVM prediction model demonstrated superior performance with an average Spearman's rank correlation coefficient of 0.9109 and root mean square error of 0.5831.
• The SVM method exhibited lower computational complexity compared to the DDT model and state-of-the-art techniques.

Conclusions:

• The proposed SVM-based virtual-view quality prediction method is effective and efficient.
• Depth distortion analysis provides a valuable basis for quality prediction in FVV systems.
• The method offers a promising solution for improving FVV system performance through accurate quality feedback.