Light Field Image Rendering With Variable View Count Adaptation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial intelligence models for rendering light field images are limited by the need for retraining when the number of views changes, and they struggle with increased time requirements for wide viewing angles.
Innovation Solution
The method involves obtaining a first light field image with a specific number of views, transforming it into a second image with a different number of views through masking or interpolation, and using an artificial intelligence model for factorization, followed by simulation to adjust the view count, and training the model based on comparison results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the number of views in light field image is increased to achieve wide viewing angle, then the viewing angle coverage is improved, but the time required to obtain input light field image increases
Solution Approach 1:
The system performs preliminary actions by capturing multiple viewpoint images in advance and storing them. When rendering is needed, the pre-captured images are directly utilized without requiring real-time capture, thus reducing the time required while maintaining wide viewing angle coverage.
Solution Approach 2:
The system creates copies of light field images captured from different viewpoints and stores them in a library. During rendering, appropriate copies are selected and processed rather than capturing new images, which significantly reduces acquisition time while preserving the ability to provide wide viewing angles.
2Measurement precision
If artificial intelligence model is retrained when number of views changes, then the model accuracy is improved, but the training time and computational resources increase
Solution Approach 1:
The system changes parameters by adjusting the number of views used from the pre-captured light field image library rather than retraining the model. The AI model processes different numbers of views from the existing training data, maintaining accuracy while avoiding retraining time and computational overhead.
3Productivity
If fixed number of views is used for light field image, then the processing speed is improved, but the adaptability to different viewing angles deteriorates
Solution Approach 1:
The system implements dynamics by allowing the number of views to be dynamically adjusted based on rendering requirements. The AI model can process a variable number of views from the pre-captured library, enabling both fast processing (when fewer views suffice) and wide viewing angle coverage (when more views are needed), thus resolving the contradiction between speed and adaptability.
Data Source
AI summary
Provided is an electronic device, wherein the at least one processor is configured to execute the at least one instruction to obtain a first light field image including view images with a first number of views, obtain, from the first light field image, a second light field image including view images with a second number of views, obtain first location information corresponding to each of sub-pixels in the second light field image, obtain a first layer image by inputting the second light field image and the first location information to an artificial intelligence model configured to perform factorization, obtain a third light field image including view images with a third number of views by inputting the first layer image to a simulation model, and train the artificial intelligence model, based on a result of comparing the first light field image and the third light field image with each other.