AI Layer Stack Shifting for Glasses-Free 3D Depth
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Glasses-free stereoscopic display systems face limitations in expressible depth range and image quality, often resulting in deteriorated images and artifacts when displaying layer stacks obtained through non-negative tensor factorization or metric factorization models.
Innovation Solution
An electronic apparatus and method that utilize an artificial intelligence model, such as a deep neural network, non-negative tensor factorization, or non-negative metric factorization model, to generate and display layer stacks with shifting parameters indicating depth information, allowing for improved depth expression and image quality by reconstructing light field images and training the model based on loss functions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If non-negative tensor factorization or non-negative metric factorization models are used to generate layer stacks, then the glasses-free stereoscopic display system can provide depth information, but the image quality deteriorates and artifacts are generated
Solution Approach 1:
The patent introduces shifting parameters as an intermediary element between the factorization model output and the final layer stack. These shifting parameters adjust the position of each layer to accurately represent depth information while preventing artifacts. The shifting operation serves as a mediator that reconciles the depth representation need with the image quality requirement.
Solution Approach 2:
The patent applies shifting parameters that modify the position coordinates of pixels in each layer of the layer stack. By changing the positional parameters of layers according to depth information, the system achieves accurate depth representation without degrading image quality. The shifting operation transforms the layer positions while preserving image fidelity.
2Ease of manufacture
If traditional factorization models are used for image rendering, then the processing method is established, but the expressible depth range is limited
Solution Approach 1:
The patent introduces dynamic shifting parameters that can be adjusted for each layer in the layer stack. Instead of using fixed layer positions, the system dynamically shifts each layer by a specific amount to achieve the desired depth effect. This dynamic approach expands the expressible depth range while maintaining the simplicity of the factorization processing method.
Solution Approach 2:
The patent segments the depth representation into multiple independent layers, each with its own shifting parameter. By dividing the depth information across multiple adjustable layers, the system can express a wider range of depths compared to traditional single-layer approaches. Each layer can be independently shifted to represent different depth planes.
3Device complexity
If layer stacks are generated without shifting parameters, then the factorization process is simple, but depth information accuracy is insufficient
Solution Approach 1:
The patent performs preliminary shifting operations on each layer before final composition of the layer stack. By pre-calculating and applying the appropriate shifting parameters to each layer, the system ensures accurate depth information is embedded in the layer structure. This preliminary action maintains the simplicity of the factorization process while improving depth accuracy.
Data Source
AI summary
An electronic apparatus includes a stacked display including a plurality of panels, and a processor configured to obtain first light field (LF) images of different viewpoints, input the obtained first LF images to an artificial intelligence model for converting an LF image into a layer stack, to obtain a plurality of layer stacks to which a plurality of shifting parameters indicating depth information in the first LF images are respectively applied, and control the stacked display to sequentially and repeatedly display, on the stacked display, the obtained plurality of layer stacks. The artificial intelligence model is trained by applying the plurality of shifting parameters that are obtained based on the depth information in the first LF images.