Research On Cracking Method Of Slider Captcha

Captcha is a major defense mechanism of the Internet today to deal with online attacks.The Dynamic Cognitive Game(DCG)Captcha is a promising new type of Captcha.It requires users to perform a series of game-type recognition.Knowing the task to pass verification has greatly improved the user experience compared to text verification and image verification.The slider-type Captcha,as a representative dynamic recognition game Captcha,requires the user to drag the slider to the target position.Compared with the naked eye recognition character or picture,the interaction process of this verification method is more interesting.Therefore,on many mainstream platforms,slide verification gradually replaces the previous popular character and image verification methods.This paper makes a detailed exploration of the current popular slider verification method,and finds that the current slide verification method can distinguish the man-machine based sliding trajectory can be artificially generated.Therefore,we propose a method of using machine learning to simulate the trajectory simulation.This will provide reference for the promotion and optimization of network information security strategies and provide a reference for the construction of a safe and harmonious internet information ecological environment.The general process adopted in the current scheme is divided into two steps: "identify the target position" and "generate the sliding track." For the former,a shape matching method can be adopted.For the special scenes verified by the slider,we have made improvements such as limiting the position,removing scaling,and rotating.For the latter,this article attempts to use the following solutions to compare the effects:(1)The database method creates a database with a large number of artificially-dragging trajectories.After parsing out the target location,the corresponding trajectories are directly extracted from the database for simulation;2)Curve fitting method,analyzing the law of artificial drag trajectory,simulating the drag trajectory with the idea of piecewise function curve;(3)Deepening reinforcement learning method,designing Deep Q Network(DQN)model,letting it Automatically generate the path;(4)Try to use Recurrent Neural Networks(RNN)to learn the sliding sequence model using the existing sliding trajectory.