Neural Face Morphing with Blended Identity Parameters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computer-generated animation and facial image manipulation technologies struggle to smoothly and temporally consistently morph faces between identities while preserving key aspects of the original identity, such as gender or ethnicity, without perceptible artifacts.
Innovation Solution
A face-morphing model is trained using a shared set of trainable neural-network parameters and identity-specific parameters, allowing for the inference of blended facial images by combining interpolation parameters and autoencoders to generate seamless transitions between input identities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional face morphing methods are used, then face transformation between identities can be achieved, but the morphing process produces perceptible artifacts and lacks temporal consistency
Solution Approach 1:
The face morphing process is segmented into distinct processing stages: feature extraction, identity encoding, blending parameter application, and image synthesis. Each stage is handled by specialized neural network components that process specific aspects of the transformation independently, allowing for better control over temporal consistency and artifact reduction in each segment of the pipeline.
Solution Approach 2:
The patent introduces intermediary representations including latent space embeddings and intermediate blended frames that serve as mediators between source and target identities. These intermediaries enable smooth transitions by providing controlled transformation pathways that maintain temporal consistency and minimize perceptible artifacts during the morphing process.
2Adaptability or versatility
If face morphing transforms between different identities, then identity characteristics change, but key aspects of the original identity such as gender or ethnicity may be lost
Solution Approach 1:
The patent applies local quality control by treating different facial features and identity characteristics with different blending weights. Key identity aspects such as gender and ethnicity are identified and protected with higher fidelity weights, while other features allow greater transformation flexibility. This enables selective preservation of important identity characteristics during morphing operations.
Solution Approach 2:
The system uses adjustable blending parameters that control the degree of transformation for different identity characteristics. By dynamically modifying these parameters based on the desired outcome and importance of various identity aspects, the system can transform between identities while preserving key characteristics that should remain constant throughout the morphing process.
3Device complexity
If a simple morphing model is used, then the system complexity is reduced, but the ability to achieve smooth and temporally consistent morphing is compromised
Solution Approach 1:
The patent combines multiple specialized neural network components into an integrated face morphing system. The encoder, blender, and decoder networks are merged into a cohesive architecture that processes images through coordinated stages, achieving high-quality temporal consistency and smooth morphing while managing system complexity through modular design and shared computational resources.
Data Source
AI summary
A method morphs an input image depicting a face to an output image depicting a face that is a blend of characteristics of a plurality of input entities. The method comprises: training a face-morphing model comprising: a shared set of parameters shared between the input identities; and, for each of the input entities, an identity-specific set of parameters. The method also comprises: receiving an input image depicting a face of one of the plurality of input identities; receiving a set of interpolation parameters; combining the identity-specific sets of trained neural-network parameters for the plurality input identities based on the interpolation parameters, to thereby obtain a blended set of neural-network parameters; and inferring an output image depicting a face that is a blend of characteristics of the input entities using the shared set of trained neural-network parameters, the blended set of neural-network parameters and the input image.