Single-Image 3D Face Modeling With Asymmetry Preservation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for generating 3D models from 2D images are either expensive or impractical, failing to accurately capture object asymmetries, particularly in applications requiring high accuracy like biometrics and medical device fitting.
Innovation Solution
A method using an artificial neural network to synthesize a pair of 2D multiview images from a single frontal image, preserving object asymmetry, which can be used to generate a 3D model with accurate facial features, suitable for biometric security and medical device fitting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional 3D scanning techniques are used to generate accurate 3D models, then manufacturing precision and measurement precision are improved, but device complexity and cost increase
Solution Approach 1:
The patent creates synthetic 2D multiview images as copies of the original 2D image from different virtual camera positions. These synthetic images serve as substitutes for actual captured images from multiple physical cameras, enabling 3D reconstruction without complex multi-camera equipment. The copying principle allows the system to generate multiple perspective views from a single image input, resolving the contradiction between accuracy and device complexity.
Solution Approach 2:
The patent introduces an artificial neural network as an intermediary between the single 2D input image and the 3D model generation process. The neural network synthesizes intermediate multiview images that bridge the gap between the single input image and the multiple views required for accurate 3D reconstruction. This intermediary approach enables accurate 3D modeling without requiring direct capture from multiple complex scanning devices.
2Measurement precision
If multiple 2D multi-view images are captured to generate accurate 3D models, then measurement precision is improved, but ease of operation deteriorates
Solution Approach 1:
The patent segments the complex task of capturing multiple 2D images from different angles into a single image capture operation. By processing a single frontal image through an artificial neural network that generates synthetic multiview images, the system eliminates the need for users to physically position and capture multiple images. This segmentation of the capture process from the processing process dramatically improves ease of operation while maintaining measurement precision through the neural network's synthesis capabilities.
Solution Approach 2:
The system performs self-service by automatically generating the required multiview images from the single captured image without requiring user intervention for additional image captures. The artificial neural network autonomously creates the synthetic views needed for accurate 3D reconstruction, eliminating the operational burden on users to capture multiple images manually. This self-service approach resolves the contradiction by making the process as easy as taking a single photo while achieving the precision of multiple views.
3Reliability
If asymmetric features are preserved in 3D models for biometric applications, then reliability is improved, but measurement precision requirements increase
Solution Approach 1:
The patent explicitly addresses asymmetry by training the artificial neural network to preserve asymmetric features during the synthesis of multiview images. The system uses asymmetric loss functions and training data that emphasize the preservation of asymmetric characteristics in facial features. By deliberately incorporating asymmetry preservation into the synthesis process, the system achieves reliable biometric measurements without requiring excessively high measurement precision, as the asymmetric features are naturally maintained through the learning process rather than demanding ultra-precise measurement.
Data Source
AI summary
An embodiment provides a method of producing a three-dimensional (3D) model of an object based on a single, frontal input two-dimensional (2D) image. In one example a method includes obtaining an actual, frontal 2D image of an object and generating a pair of synthetic 2D multiview images of the object based on the actual, frontal 2D image of the object. A 3D model of the object is produced based on at least the pair of synthetic 2D multiview images. The 3D model conserves an asymmetry of the object. An output using the 3D model of the object is produced that conserves the asymmetry.