Geodesic Intra-Prediction for Spherical Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional intra-frame prediction mechanisms in video coding are inefficient for 360-degree or spherical video due to the distortion of straight lines into curved lines during projection, leading to increased residual information and bit rate.
Innovation Solution
The solution involves adjusting intra-prediction to align with the geometry of spherical video by performing intra-prediction along geodesic curves, which correspond to straight lines in the 3D scene, thereby reducing the need for signaling curved lines as sets of small straight lines.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If conventional intra-frame prediction is performed along straight lines, then the prediction mechanism is simple and straightforward, but the prediction efficiency deteriorates for spherical video due to line distortion during projection
Solution Approach 1:
The patent applies curvature by transitioning from straight line prediction to geodesic curve prediction on the spherical surface. The geodesic curves account for the spherical geometry of 360-degree video, allowing prediction to follow the natural curvature of the sphere rather than forcing straight lines that distort during equirectangular projection. This resolves the contradiction by maintaining prediction simplicity through a unified curved approach rather than attempting to represent curves as sequences of straight lines.
2Power
If straight lines are used for intra-prediction, then the prediction process is computationally efficient, but the amount of residual information increases due to mismatch with curved projection lines
Solution Approach 1:
The patent reduces residual information by aligning the prediction paths with the actual geometry of the projected spherical video. By using geodesic curves that correspond to straight lines on the sphere before projection, the prediction accurately follows the curved appearance in the equirectangular projection, thereby minimizing prediction errors and residual information while maintaining computational efficiency through a unified mathematical approach.
3Device complexity
If conventional directional prediction modes are used, then the bitstream signaling is straightforward, but the prediction accuracy deteriorates for curved projection lines
Solution Approach 1:
The patent improves prediction accuracy by replacing straight line prediction modes with geodesic curve prediction modes that naturally follow the curved projection of spherical video. The geodesic curves are defined by their starting point, ending point, and curvature, which can be signaled in the bitstream. This approach maintains relatively simple signaling while dramatically improving prediction accuracy for spherical video content.
Data Source
Figure 1A~1B
Figure 2A~2B
Figure 2C~2D
AI summary
According to a first aspect a video encoder is provided. The video encoder comprises an input unit that is configured to receive frames of spherical video, each of the frames comprising blocks of pixels. The video encoder further comprises an intra-prediction unit that is configured to generate a set of residuals for a current block to be encoded, by performing intra-prediction along a geodesic curve for the current block to be encoded. The video encoder further comprises an output unit that is configured to provide an encoded bitstream based on sets of residuals generated by the intra-prediction performed on the blocks to be encoded.