An inverse face recognition method based on PSO

Abstract

The invention discloses an anti-face recognition method based on PSO, comprising the following steps: S1, pre-processing the attacker's face data and adding a data set; 2, training a face classifier by use a mixed data set; 3, setting PSO parameters and antagonizing attack parameters; S4: initializing the PSO and starting the iterative evolution process to find the optimal countermeasure against the disturbance; S5: extracting and materializing the countermeasure accessories for physical attack test. The invention aims at the black box face recognition system, generates the optimal anti-disturbance through the PSO evolutionary strategy, realizes the materialization of the anti-disturbance through the facial accessories, can realize the effective anti-face recognition in the digital and physical environment, and has better model generalization ability and practical application value compared with other white box attack strategies.

Description

technical field [0001] The invention relates to the fields of computer vision and machine learning, in particular to a PSO-based anti-face recognition method. Background technique [0002] With the rapid development of artificial intelligence, face recognition research methods have evolved from early template matching, PCA principal component analysis to manual feature extraction, and then to today's mainstream deep learning. This technology has matured and been successfully applied to the market. [0003] However, researchers have found that neural networks have natural flaws, and adversarial samples generated through adversarial attack strategies will interfere with the accuracy of deep learning models on tasks such as image recognition. The attacker generates a data sample with adversarial perturbation through the adversarial attack strategy, and inputting the adversarial sample into the neural network will cause the model to change its prediction, which is reflected in t...

AI Technical Summary

Problems solved by technology

[0006] In order to overcome the problems that existing face recognition adversarial attacks ignore the physical realizability of adversarial disturbances and the generalization ability of attack strategies, etc., the present invention provides a PSO-based anti-face recognition method, which not only considers the black box model Simulates real scenes while also utilizing facial accessories for physicalization against perturbations

Images