Face Recognition Synthetic Image Registration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face recognition methods either increase recognition execution time and operational burden by training deeper neural networks or require multiple user images, leading to a poor user experience.
Innovation Solution
A face recognition system and method that generates synthetic images based on adjustment parameters, compares their features with the original image's features, and registers them if the similarity is below a certain threshold, thereby increasing recognition success rate while maintaining user experience.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If more different face images are received to establish samples, then the success rate of face recognition is increased, but the user experience deteriorates
Solution Approach 1:
The patent generates synthetic face images by copying and transforming the user's original face image through various synthesis operations (rotation, scaling, translation, filtering). These synthetic images serve as additional training samples without requiring the user to provide multiple real images, thus improving recognition success rate while maintaining good user experience.
2Reliability
If a deeper neural network is trained to increase recognition success rate, then the recognition accuracy is improved, but the execution time and operational burden increase
Solution Approach 1:
The patent performs face image synthesis and sample generation in advance during the registration phase. By pre-generating synthetic training samples and training the neural network beforehand, the system avoids performing complex computations during actual recognition, thus reducing recognition execution time while maintaining high success rate.
Solution Approach 2:
The patent transforms the original face image using various synthesis parameters (rotation angles, scaling factors, translation distances, filtering parameters) to generate diverse synthetic images. These parameter changes create varied training samples that improve recognition accuracy without requiring a deeper neural network architecture.
3Reliability
If synthetic images are generated and registered to increase recognition success rate, then the recognition accuracy is improved, but the system complexity increases
Solution Approach 1:
The patent introduces a face image synthesis module as an intermediary between the user's original image and the training dataset. This module generates synthetic images that bridge the gap, providing diverse training samples without requiring complex data collection processes or multiple user inputs, thus improving accuracy while controlling system complexity.
Data Source
AI summary
A face recognition system and a method for enhancing face recognition are provided. The method includes: receiving a face image and obtaining a feature of the face image from a feature extraction model; registering the face image to set the feature of the face image as a first recognition feature; performing a synthesis operation on the face image according to at least one first adjustment parameter to generate a synthetic image, and obtaining a feature of the synthetic image from the feature extraction model; comparing first recognition feature with the feature of the synthetic image to obtain a feature similarity; comparing the feature similarity with a threshold value to obtain a comparison result; and registering the synthetic image when the comparison result indicates that the feature similarity is less than or equal to the threshold value.