Biometric Identification Training Data Selection by Embedding Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The creation and use of biometric training data for embedding models is costly due to the expense of obtaining actual biometric input data and the computational resources required for generating synthetic data, while existing synthetic data may not accurately represent actual individuals.
Innovation Solution
A system and method to select and utilize synthetic biometric input data as a proxy for unenrolled users and generate data as needed, incorporating a quality assessment module to improve the training process, and strategically add synthetic data to the training set based on proximity to query embedding data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If actual biometric input data is obtained for training, then training accuracy is improved, but cost and time required for data collection increase
Solution Approach 1:
The patent uses synthetic biometric data as copies or representations of actual biometric data. These synthetic data are generated to resemble real biometric patterns while avoiding the need to collect actual data from individuals, thus reducing time and cost while maintaining training effectiveness through the use of realistic but fictional biometric samples
2Loss of time
If synthetic biometric data is generated for training, then cost and time for data collection are reduced, but the accuracy of representing actual individuals deteriorates
Solution Approach 1:
The patent transforms actual biometric data into synthetic data by changing parameters such as modifying pixel values, adjusting geometric characteristics, or altering feature distributions while preserving the essential patterns. This allows the data to represent real individuals accurately enough for training while being sufficiently different to avoid privacy issues and reduce collection time
Solution Approach 2:
The system creates synthetic biometric data that copies the statistical properties and structural characteristics of actual biometric data. By learning from real data and generating synthetic versions that replicate key features, the system maintains representational accuracy while eliminating the need for direct collection from individuals
3Measurement precision
If more training data is used to improve accuracy, then training precision is improved, but computational resources and time required increase
Solution Approach 1:
The patent changes the parameters of training data by transforming actual biometric data into synthetic data with modified characteristics. This allows the system to maintain training precision while reducing the computational burden of processing large volumes of real data, as synthetic data can be generated efficiently and scaled without proportionally increasing computational costs
4Use of energy by moving object
If existing synthetic data is used for training, then computational resources are reduced, but the quality and representativeness of training data deteriorates
Solution Approach 1:
The patent incorporates feedback mechanisms where the system evaluates the quality of synthetic training data and adjusts the generation process accordingly. By continuously monitoring training performance and data representativeness, the system refines synthetic data generation to improve quality while maintaining computational efficiency
Solution Approach 2:
The system performs preliminary actions by pre-processing and preparing actual biometric data to serve as templates for generating synthetic data. This preliminary preparation ensures that the synthetic data will be of high quality and representative, avoiding the need for extensive computational resources during the main training process
Data Source
AI summary
A biometric identification system acquires an image of at least a portion of a user that is processed by an embedding model. Additional training of the embedding model is performed using selected training data. In one implementation, the selected training data is determined based on similarity to embedding data acquired during operation. For example, a query image is processed to determine query embedding data. If the query embedding is deemed to be a close match to embedding data of a previously enrolled user, additional synthetic training data that is also similar to the query embedding data is selected for use in subsequent training. In another implementation, enrolled user data may be analyzed to find close matches to synthetic training data, from which selected training data is determined. Use of the additional synthetic training data facilitates the subsequent training in distinguishing between similar instances of embedding data.