Adversarial sample generation method and device, medium and computing device

Abstract

The invention provides an adversarial sample generation method, which comprises the following steps of obtaining a first sample image which is an image of an attack object; based on the first sample image, solving by taking the mathematical expectations of a similarity function of an identification object of a preset white box substitution model under different shooting conditions as an objective function, and generating an interference image of a specific area; and superposing the interference image on the first sample image to generate an adversarial sample. According to the present invention, the white box substitution model is used to generate the adversarial sample, so that the attack efficiency can be improved, whether the potential vulnerabilities exist in an image recognition model or not can be rapidly detected, the evaluation of the image recognition model is improved, and the safety of the image recognition model is improved. In addition, the embodiment of the invention provides an adversarial sample generation device, a medium and a computing device.

Description

technical field [0001] The present invention relates to the technical field of computer vision, in particular to a method, a device, a medium and a computing device for generating an adversarial sample. Background technique [0002] As an important task in computer vision, image recognition has also achieved tremendous development driven by deep neural networks. And image recognition systems have many applications in real-life scenarios such as finance / payment, public transportation, and criminal identification. Although image recognition systems have achieved great success and practical applications, these systems cannot fully ensure sufficient security. [0003] In recent years, deep learning has achieved breakthroughs in the fields of image, speech and natural language. However, for some deep neural network models that can achieve a high accuracy recognition rate, they are also vulnerable to attacks in an adversarial environment. In an adversarial environment, the deep...

AI Technical Summary

Problems solved by technology

[0004] At present, most of the image recognition systems on the market are integrated devices that are packaged in black boxes and only provide camera input. The specific model structure and parameters cannot be known, and the security of such systems is difficult to verify.

Images