Mobile 3D Face Model Generation via CNN Parameter Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Generating a three-dimensional (3D) face model from a two-dimensional (2D) image on a mobile device is challenging due to the complexity of algorithms requiring high-performance hardware, making it difficult to implement without additional hardware support.
Innovation Solution
A method using a trained convolutional neural network (CNN) to obtain camera and face model parameters from a 2D image, generating a 3D face model by applying these parameters to a 3D morphable face model, and overlaying the model on the 2D image, which reduces resource intensity and allows for generation on mobile devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional 3D face model generation algorithms are used, then measurement precision is improved, but device complexity increases and requires high-performance hardware
Solution Approach 1:
The patent transforms the complex 3D face model generation problem into a parameter estimation problem. Instead of using traditional complex algorithms, the system estimates key parameters (landmark positions, face orientation angles, scale) from 2D images and uses these parameters to construct 3D face models. This parameter-based approach significantly reduces computational complexity while maintaining accuracy.
Solution Approach 2:
The patent replaces traditional mechanical/computational 3D reconstruction algorithms with a machine learning-based parameter estimation system. The neural network learns to directly predict 3D face parameters from 2D images, substituting complex iterative reconstruction algorithms with a more efficient learned model that runs on mobile devices.
2Measurement precision
If traditional 3D face model generation algorithms are used, then measurement precision is improved, but ease of operation worsens due to hardware requirements
Solution Approach 1:
The patent changes the computational approach from full 3D reconstruction to parameter estimation. By focusing on estimating essential parameters (landmarks, orientation, scale) rather than computing complete 3D geometry, the system reduces computational burden to levels suitable for mobile devices while preserving the ability to generate accurate 3D face models when needed.
Solution Approach 2:
The patent segments the 3D face model generation process into independent parameter estimation tasks. The system separately estimates landmarks, face orientation, and scale parameters, then combines them to construct the 3D model. This segmentation allows each sub-task to be optimized for mobile computation while maintaining overall accuracy.
3Productivity
If additional hardware support is provided, then productivity is improved, but device complexity increases
Solution Approach 1:
The patent enables mobile devices to perform 3D face model generation using their own existing resources (camera, processor, memory) without requiring additional specialized hardware. The optimized parameter estimation algorithm is designed to run efficiently on standard mobile device components, making the system self-sufficient.
Solution Approach 2:
The patent changes the computational parameters of the algorithm to match mobile device capabilities. By reformulating the 3D reconstruction problem as a parameter estimation task with reduced computational requirements, the system achieves practical processing speeds on mobile devices without needing additional hardware acceleration.
Data Source
AI summary
Provided is a two-dimensional (2D) image processing method including obtaining a 2D image, processing the obtained 2D image by using a trained convolutional neural network (CNN) to obtain at least one camera parameter and at least one face model parameter from the 2D image, and generating a three-dimensional (3D) face model, based on the obtained at least one camera parameter and at least one face model parameter.