Iris image feature extraction method, system and device based on federated learning

Abstract

The invention belongs to the field of image recognition, particularly relates to an iris image feature extraction method, a system and a device based on federated learning, and aims to solve the problems that an existing iris feature extraction model cannot give consideration to privacy protection and multi-party cooperation, and the recognition performance of the model cannot be improved by purely stacking iris data for model training. According to the invention, the local iris features of the local iris data set are obtained, so that the local feature triple and the local ternary loss are generated; a Wasserstein federation ternary loss function is calculated at a third party, an iris image feature extraction network of the local platform is updated according to the Wasserstein federation ternary loss function, and new iris image features are extracted by using the new iris image feature extraction network. According to the method, distribution characteristics and rules of iris features from different sources in the feature space are utilized to assist model training, so that the extracted features have better distinguishability, and therefore, the iris feature expression ability and recognition accuracy of each partner are improved.

Description

technical field [0001] The invention belongs to image recognition, and in particular relates to a federated learning-based iris image feature extraction method, system and device. Background technique [0002] As a biometric feature with high accuracy and good security, the iris is being widely used in various identification and security control scenarios. After strict, it is not feasible to build a large-scale iris data center for stack training. Therefore, it is imperative to establish a federated iris recognition model training and feature learning framework based on privacy protection and multi-party cooperation. Federated learning can combine multi-party data for training, which is more suitable for occasions with high security and privacy requirements. On the other hand, the federated iris recognition model training and feature learning framework can combine multiple iris databases and various iris recognition neural networks for collaborative learning, fully balance ...

AI Technical Summary

Problems solved by technology

[0003] In the existing technology, by collecting multiple iris data sets, without considering data security and privacy protection, a single data center is established, all data are directly stacked and mixed together, and sent to models such as deep neural networks for feature learning training Way

[0004] The iris recognition model based on the deep learning framework usually requires a large number of iris samples for learning, but the construction of a large-scale iris data center has security risks and privacy protection limitations, because the original biometric data, which is highly private, will be irreversible once leaked. However, the generalization ability and marginal effect of the iris recognition model lacking a large amount of training data will be greatly reduced in practical applications. Therefore, it is particularly important to establish a framework for joint multi-party data decentralized iris recognition model training and feature learning.

In addition, most of the current iris data sets have obvious differences in collection rules, collection equipment, collection environment, data scale, image quality, etc., and the distribution of different iris databases is significantly different, simply relying on stacking unlimited irises The data cannot continuously improve the recognition performance of the model, and may even have a counterproductive effect

At present, there is a lack of an iris recognition model training and feature learning framework that takes into account privacy protection and multi-party cooperation to fully tap the potential and deep value of multi-party data, so that the recognition models of each partner can benefit from the spatial distribution of other data sets and model feature expression ability, so as to further improve its recognition ability and generalization ability

Images