Confrontation learning-based method and system for generating image confrontation verification codes

Abstract

The invention discloses a confrontation learning-based method for generating image confrontation verification codes. The method comprises the following steps of: (1) selecting an image recognizing neural network A and an attack algorithm a; (2) generating an image confrontation verification code on the basis of a confrontation learning method according to the trained image recognizing model A andthe attack algorithm; (3) selecting one or more image recognizing models B, selecting one or more attack algorithms b, cracking the image confrontation verification code generated in the step (2) according to the image recognition model B and the attack algorithm b and adjusting a generation strategy of the image confrontation verification code according to the cracking result; and (4) generatingan image confrontation verification code according to the adjusted generation strategy of the image confrontation verification code. The invention also discloses a system for generating the image confrontation verification codes. The method disclosed by the invention has the beneficial effects that a small amount of special noise is added to an image verification code through confrontation learning to attack a depth image recognizing model and the security of the image verification code is protected in the manner of weakening the attack ability of a verification code cracker.

Description

technical field [0001] The invention relates to the fields of machine learning model security and image verification code security, in particular to an image confrontation verification code generation method and system based on confrontational learning. Background technique [0002] Captcha is a Turing test that is easy for humans to solve but hard for current programs to solve. It is widely deployed on Web pages, and can effectively prevent some malicious behaviors, such as batch registration of zombie accounts, hacking, etc. [0003] There are many types of verification codes, and one of the commonly used verification codes is a verification code based on image recognition, referred to as an image verification code. Image captchas require users to correctly select all images in a set of images that match a certain semantic meaning. Compared with other types of verification codes, image verification codes have a better balance between security and usability: images contai...

AI Technical Summary

Problems solved by technology

However, in the verification code attack and defense environment, we, as the defender, cannot know the model parameters used by the verification code cracker in advance; even the cracker will use image preprocessing methods to process the image before recognizing the image to remove noise

In this way, adding noise to the image cannot achieve the effect of deceiving the cracker's classifier, and the existing noise generation method based on adversarial learning cannot meet the needs of anti-captcha defense

Images