OMECNN:一种基于有序马尔可夫枚举器和判别神经网络的口令生成模型

OMECNN: a password-generation model based on ordered markov enumerator and critic neural network

摘要:

基于口令的身份鉴别是目前最流行的鉴别方式之一，利用口令生成技术进行大规模口令集的生成，进而检测现有用户口令保护机制的缺陷、评估口令猜测算法效率等，是研究口令安全性的重要手段.本文提出一种基于有序马尔可夫枚举器和判别神经网络的口令生成模型OMECNN, 使用有序马尔可夫口令枚举器按照口令组合概率的高低生成组合口令，同时基于判别神经网络进行打分筛选口令，选出得分高于阈值的口令组成最终口令集.采用本文提出方法生成的口令集具有按照口令组合概率高低排序的特点，以及符合真实训练口令集的口令分布的特点.实验结果表明，在生成1e7条口令时，OMECNN模型生成的口令集在Rockyou测试集上的匹配条目比OMEN模型高出16.60%，比PassGAN模型高出220.02%.

Abstract:

Password identification is one of the most popular way of identification. Generating a large-scale password set based on password-generation techniques is a principal method to research password security, which can be applied to evaluate the efficiency of password-generation algorithm and detect the defects of existing user-password protective mechanisms. In this paper, we propose a password-generation model based on an ordered Markov enumerator and critical neural network (OMECNN). The OMECNN model combines both Markov chain and neural network techniques. OMECNN utilizes the ordered Markov passwords enumerator to generate the passwords according to the probability of combinations, and then uses the critic neural network to score those passwords, and selects the passwords whose score is higher than the threshold to form the final password set. The generated password set has the characteristics of sorting according to the combination probability of passwords and the distribution of passwords in accordance with the real training password set. The experimental results show that when 1e7 passwords are generated, the hits of OMECNN model on Rockyou test set is 16.60% higher than that of OMEN model and 220.02% higher than that of Pass GAN model.