360-Degree Image Projection Detection Using Pixel Stretch Cues
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image players lack the ability to accurately determine the projection type used to encode 360-degree images without user input, as projection-type metadata is often absent, and machine-learning models for this purpose are computationally expensive.
Innovation Solution
A lightweight computer-vision technique is employed to detect equirectangular projection in 360-degree images by analyzing pixel stretch factors and edge similarities, using algorithms to calculate confidence scores based on pixel differences and similarities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine-learning models are used to predict projection type, then detection accuracy is improved, but computational cost increases
Solution Approach 1:
The patent replaces expensive machine-learning models with inexpensive geometric algorithms that analyze pixel relationships. Instead of using computationally intensive neural networks, the invention employs simple mathematical operations comparing pixel values at different positions to detect equirectangular projection characteristics, achieving accurate detection with minimal computational resources
Solution Approach 2:
The patent substitutes the 'mechanical' machine-learning model system with a mathematical/computational system based on geometric relationships. The solution replaces complex AI processing with deterministic algorithms that calculate pixel differences and ratios according to known equirectangular projection geometry, eliminating the need for training and inference overhead
2Adaptability or versatility
If projection-type metadata is not included in image files, then file compatibility is improved, but detection capability deteriorates
Solution Approach 1:
The patent enables the image decoding system to automatically determine projection type through self-analysis of the image data itself. The algorithm examines intrinsic geometric properties of the image pixels without requiring external metadata, allowing the system to self-identify the projection type and configure appropriate decoding parameters
Solution Approach 2:
The patent uses pixel value variations (analogous to color changes) as indicators of projection type. By analyzing how pixel values change across the image according to known equirectangular projection patterns, the system can detect the projection type through these inherent visual signatures embedded in the image data
3Measurement precision
If user input is required to specify projection type, then detection accuracy is improved, but ease of operation deteriorates
Solution Approach 1:
The system performs automatic projection type detection without requiring user intervention. The algorithm independently analyzes the image data and determines the projection type, eliminating the need for users to manually specify or configure projection settings while maintaining accurate detection
Solution Approach 2:
The patent performs projection type detection automatically during the image decoding initialization phase, before user interaction is needed. By determining the projection type in advance through automatic analysis, the system prepares the correct decoding parameters without requiring users to provide input or make selections
Data Source
AI summary
In one embodiment, a method includes accessing an image file representing a 360-degree image of a scene and determining, from the image file, a stretch factor that identifies to a relative change of adjacent pixels in at least one of (1) an upper row of pixels or (2) a lower row of pixels. The method further includes determining, from the image file, an edge similarity that identifies a change between pixels in a first column and pixels in a last column and determining, based on the stretch factor and on the edge similarity, whether the 360-degree image was encoded using an equirectangular projection and then rendering the 360-degree image based on the equirectangular-projection encoding.