Projection-Based Image Decoding for 360-Degree Compression Bottlenecks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems struggle with the massive data generated from processing multi-view images for 360-degree images in virtual and augmented reality, leading to insufficient performance in encoding and decoding high-resolution images.
Innovation Solution
A method for encoding and decoding 360-degree images that includes generating a predicted image using syntax information, combining it with a residual image, and reconstructing the image based on projection formats like ERP, CMP, OHP, and ISP, with image expansion and motion vector prediction to enhance compression performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional image processing methods are used for 360-degree images, then processing can be performed with standard tools, but the performance is insufficient for handling massive data from multi-view images
Solution Approach 1:
The patent divides the 360-degree image processing into multiple projection formats (ERP, CMP, OHP, ISP) and processes different regions with different methods. The image is segmented into face regions and non-face regions, with motion compensation applied differently to each segment, enabling efficient handling of large data volumes through localized processing strategies
Solution Approach 2:
The patent transforms the processing approach by introducing projection format dimensionality conversion. By converting between different projection formats (e.g., from equirectangular to cube map), the system can leverage the strengths of each format for specific processing tasks, improving overall processing performance for massive datasets
2Manufacturing precision
If high-resolution encoding is performed, then image quality is improved, but the compression efficiency decreases leading to larger data volumes
Solution Approach 1:
The patent applies different quality levels and processing methods to different regions of the image. Face regions receive full motion compensation and high-quality processing, while non-face regions use simplified methods. This local quality differentiation maintains perceptual image quality while significantly improving compression efficiency by reducing redundant data in less important regions
Solution Approach 2:
The patent dynamically adjusts processing parameters based on region type and motion characteristics. Motion vector precision, block size, and compensation methods are changed according to the specific region being processed, enabling optimal balance between quality and compression ratio for different parts of the image
3Measurement precision
If motion compensation is applied to all regions, then prediction accuracy is improved, but the processing complexity and time increase
Solution Approach 1:
The patent segments the image into face regions and non-face regions, applying motion compensation only to face regions where it provides significant benefit. This segmentation maintains prediction accuracy for important regions while reducing overall processing complexity by eliminating unnecessary compensation calculations in non-face regions
Solution Approach 2:
The patent applies motion compensation selectively rather than universally. By performing motion compensation only on face regions and using simpler methods for non-face regions, the system achieves sufficient prediction accuracy for perceptually important areas while avoiding the excessive processing complexity that would result from applying full motion compensation to the entire image
Data Source
AI summary
A method of decoding an image, includes obtaining at least one offset for a picture, deriving a variable for scaling for the picture based on the at least one offset, and performing inter prediction based on the variable for scaling for the picture. The at least one offset is defined with a direction of scaling.