Facial recognition optimization based on adversarial sample generation in the field of artificial intelligence

Facial recognition systems in the artificial intelligence have potential threats from sample attacks and pose security risks. Thus, the research on facial adversarial sample generation based on ensemble learning is carried out. Firstly, the traditional AdaBoost is improved using particle swarm optimization algorithm and dual threshold classification method. Then, an integrated learning facial adversarial sample generation algorithm based on improved AdaBoost is designed. The results show that the improved AdaBoost algorithm can significantly improve training speed while ensuring algorithm performance, with an average training time of 53 seconds and fewer average weak classifiers. The sample recognition rate of the proposed improved AdaBoost algorithm is 96.41%, with a positive sample misclassification rate of 3.24. The facial recognition rate on the CMU dataset is 94.7%, with 8 false positives. The integrated learning facial adversarial sample generation algorithm based on improved AdaBoost successfully attacked 716 images from the LFW dataset. Its success rate is higher than traditional attack algorithms. The structural similarity and peak signal-to-noise ratio of the proposed algorithm are 0.91 ± 0.02 and 29.81 ± 4.07, respectively, and the learned perceptual image block similarity is 0.02 ± 0.008. The experimental results demonstrate that the samples generated by the proposed algorithm have high similarity and good perceptual quality. In summary, the algorithm designed in this study has good application effects in generating facial adversarial samples, promoting the advancement of facial recognition technology.