3D Image Generation from Single Video and Photo

The patent uses image synthesis technology to generate synthetic images that replicate the appearance of multiple gaze-position images. Instead of requiring users to prepare multiple actual images or videos, the system creates virtual copies through deep learning models that synthesize appropriate images based on a single input image and gaze position data, thereby maintaining 3D image provision capability while dramatically improving ease of operation

The patent changes the fundamental parameter from requiring multiple source images/videos to using only a single source image. By utilizing gaze position information as a controlling parameter and applying image synthesis algorithms, the system transforms one input image into multiple gaze-specific output images, resolving the contradiction between adaptability and ease of operation

The patent replaces the mechanical approach of physically preparing and storing multiple images or videos with a computational image synthesis system. Instead of requiring a complex library of pre-prepared multi-gaze images, the system uses algorithms and deep learning models to generate images on-demand from a single source image and gaze position data, thereby reducing device complexity while maintaining stereoscopic image quality

The patent performs preliminary action by extracting facial features and preparing a single source image in advance, then uses this pre-processed data along with gaze position information to generate multiple stereoscopic images. This preliminary preparation simplifies the overall system complexity by avoiding the need to manage and process multiple source images simultaneously