Deep Learning Hologram Synthesis for Real-Time 3D Display
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional computer-generated hologram (CGH) methods face challenges in real-time image processing and optical reconstruction due to high computational load and data requirements for producing 360° video content, limiting interactive experiences for users.
Innovation Solution
A deep learning method is employed to generate holographic 3D data from light field refocus images using a convolutional neural network (CNN), enabling high-speed hologram content production and interaction by integrating user voice and gesture recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional computer-generated hologram (CGH) methods are used to synthesize digital hologram content from 3D data, then holographic content can be produced, but the computational load becomes excessively high and real-time image processing cannot be achieved
Solution Approach 1:
The patent replaces conventional mathematical computation methods (FFT-based algorithms) with a deep learning model (neural network) to perform hologram synthesis. This substitution transforms the computational mechanism from traditional signal processing to AI-based inference, dramatically reducing computational load and enabling real-time hologram generation from RGB-D inputs.
2Loss of time
If conventional CGH methods process 3D data to produce 360° video content, then complete holographic content is generated, but the data processing time becomes a significant burden
Solution Approach 1:
The patent performs preliminary training of the deep learning model using RGB-D pairs and corresponding hologram data before actual hologram generation. This pre-training phase enables the model to learn the mapping relationship in advance, so that during real-time operation, holograms can be generated instantly from RGB-D inputs without requiring complex mathematical calculations at runtime.
3Adaptability or versatility
If conventional CGH methods are used for hologram synthesis, then digital hologram content can be produced, but interactive experiences between users and content cannot be enabled
Solution Approach 1:
The patent creates a universal deep learning model that can process various types of 3D inputs (RGB-D data from different sources, different viewpoints) and generate corresponding holograms. This multi-functional approach enables the system to handle diverse interaction scenarios (user movement, viewpoint changes, object manipulation) through a single unified model, facilitating real-time interactive experiences without requiring separate processing pipelines for each interaction type.
Data Source
AI summary
A system and method for hologram synthesis and processing capable of synthesizing holographic 3D data and displaying (or reconstructing) a full 3D image at high speed using a deep learning engine. The system synthesizes or generates a digital hologram from a light field refocus image input using the deep learning engine. That is, RGB-depth map data is acquired at high speed using the deep learning engine, such as a convolutional neural network (CNN), from real 360° multi-view color image information and the RGB-depth map data is used to produce hologram content. In addition, the system interlocks hologram data with user voice recognition and gesture recognition information to display the hologram data at a wide viewing angle and enables interaction with the user.