Adaptive Effective Wiener Filter- and Regression-based Upsampling for Asymmetric Resolution Stereoscopic Video Coding

Area of Science:

• Computer Vision
• Signal Processing
• Multimedia Systems

Background:

• Asymmetric Resolution Stereoscopic Video Coding (ARSVC) reduces bitrate by using a full-sized left view and a quarter-sized right view.
• Upsampling the downsampled right-view video is crucial for displaying 3D scenes on client devices.

Purpose of the Study:

• To develop an effective adaptive upsampling method for ARSVC.
• To improve the visual quality of upsampled stereoscopic video sequences.

Main Methods:

• Proposed Resolution- and Texture-Consistency Wiener Filter (RTCWF) for enhanced spatial domain upsampling accuracy.
• Developed Linear Regression-based Interview Prediction with Residual Compensation (LRIPRC) for improved interview domain accuracy.
• Introduced an adaptive fusion approach (RTCWFLRIPRC) integrating RTCWF and LRIPRC to maximize upsampled image quality.

Main Results:

• Experimental results in 3D-HEVC demonstrated superior performance of the RTCWF-LRIPRC method.
• The proposed method outperformed state-of-the-art upsampling techniques across six quality metrics.
• The method met execution time requirements for practical applications.

Conclusions:

• The RTCWF-LRIPRC method offers significant quality improvements for ARSVC upsampling.
• This approach effectively enhances the visual experience of stereoscopic video.
• The adaptive fusion strategy successfully integrates spatial and interview domain predictions.