GAN Face Synthesis for Facial Recognition Enrollment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Facial identification systems require numerous and diverse sample images for accurate enrollment, which is burdensome and inaccessible to average users due to their complexity and cost.
Innovation Solution
A method using a generative adversarial network (GAN) to generate facial images by maintaining the identity of the input face through feedback loops between the generator and discriminator, allowing for the creation of positive and negative galleries to train a facial recognition network with a single input image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large number of cameras with known properties are placed at known positions under carefully controlled settings to generate sample facial images, then the facial identification accuracy is improved, but the device complexity and cost increase significantly
Solution Approach 1:
The patent uses a GAN generator to create synthetic copies of facial images that resemble real captured images. These generated images serve as artificial training data, replacing the need for multiple physical cameras and controlled settings. The generator learns from real images and produces realistic synthetic variations, effectively copying the appearance and characteristics of real facial data without requiring complex physical capture systems.
2Measurement precision
If a large number of cameras with known properties are placed at known positions under carefully controlled settings to generate sample facial images, then the facial identification accuracy is improved, but the cost increases significantly
Solution Approach 1:
The patent replaces expensive, delicate camera systems with a computational approach using software-based GAN models. The synthetic images generated are effectively 'disposable' in the sense that they can be created infinitely at minimal computational cost once the model is trained. This eliminates the need for expensive physical hardware while maintaining the ability to generate diverse training data for accurate facial identification.
3Measurement precision
If a large number of cameras with known properties are placed at known positions under carefully controlled settings to generate sample facial images, then the facial identification accuracy is improved, but the ease of operation decreases significantly
Solution Approach 1:
The system performs self-service by automatically generating diverse facial images through the GAN model without requiring user involvement in the complex process of setting up multiple cameras and controlled environments. The generator autonomously creates synthetic training data from a single real facial image provided by the user, eliminating the burden of coordinating complex physical capture systems while still achieving high identification accuracy.
4Measurement precision
If more numerous and diverse sample facial images are provided during enrollment, then the facial identification accuracy is improved, but the quantity of images required increases the burden on the user
Solution Approach 1:
The patent segments the task of generating diverse facial images into two parts: a single real captured image provided by the user, and multiple synthetic variations generated by the GAN model. This segmentation allows the system to obtain numerous diverse training images (improving accuracy) without requiring the user to provide or capture all of them manually (reducing burden). The generator handles the multiplication and diversification of images computationally.
Data Source
AI summary
Training a generative adversarial network (GAN) for use in facial recognition, comprising providing an input image of a particular face into a facial recognition system to obtain a faceprint; obtaining, based on the input faceprint and a noise value, a set of output images from a GAN generator; obtaining feedback from a GAN discriminator, wherein obtaining feedback comprises inputting each output image into the GAN discriminator and determining a set of likelihood values indicative of whether each output image comprises a facial image; determining, based on each output image, a modified noise value; inputting each output image into a second facial recognition network to determine a set of modified faceprints; defining, based on each modified noise value and modified faceprint, feedback for the GAN generator, wherein the feedback comprises a first value and a second value; and modifying control parameters of the GAN generator.