Research Of Slider Game CAPTCHA And Captcha Recognition Based On Sample Reusing

In recent years, many commercial websites have used captcha techniques sequentially. As an important means to improve website security, CAPTCHA can successfully solve a range of security issues such as spam attacks, bulk registration and distributing the stolen resources, false voting on on-line voting system and brute force to obtain account passwords and so on. Now most CAPTCHAs can greatly increase the difficulty of attacks through a complex twist deformation or hollow technology, but it led to a substantial decline in the user experience, thus in danger of losing the site's users, then the behavioral CAPTCHA arises at the historic moment. However, The databases of behavior-based CAPTCHA was too small and the operating path was regular, which can not resist further robot attacks successfully. Therefore, on the one hand, in order to ensure a good user experience, while increasing the difficulty of cracking behavioral CAPTCHA, we proposed a slider game CAPTCHA, which is more difficult to recognize. Experimental results show that, compared with the general behavior-based CAPTCHA, slider game CAPTCHA is more popular with the users.On the one hand, Deep learning is now the mainstream direction of CAPTCHA recognition. but its high recognition rate depends on the mass number of samples. In order to reduce training costs, this paper presents a CAPTCHA recognition method based on a small sample with the introduction of active learning strategies, reducing the initial training set by sample multiplexing. We use the relative entropy to calculate the value of uncertainty to improve accuracy. The main contribution is shown as follows:1. For the problems of the datasets of the behavioral CAPTCHAs was too small and the operating path was regular, we proposed a new slider game CAPTCHA.The generation of the CAPTCHA is, firstly, generated a background image, adding four randomly generated characters and numbers, noise, interference lines, thereby generating a reference image, and then number the reference image randomly; subsequently, use the reference picture for the motherboard, generate multiple distorted images, and number the new generation distorted pictures.Next, Connect the duplicates to the reference image to compose a long picture in accordance with the number of each pictures, Finally, sends the long picture to the client, the user drags the slider to match the reference image,which is the correct answer, to complete the validation process. The validation method using behavior-based authentication technology, obtained higher security, can effectively prevent the attacks form machine. Even more noteworthy is, it does not require the traditional keyboard input.2. Aiming at the problems of the large sample size and the high cost, with the introduction of active learning algorithm, we proposes a CAPTCHA recognition method based on small samples, compared to the method of Fabian(98.07% recognition rate), We obtained a higher recognition rate.To sum up, the proposed Slider game CAPTCHAs can promote the combination of human body engineering and CAPTCHA generation technology; the CAPTCHA recognition based on small samples, using the sample reuse, to explore human learning method based on small sample set up new ideas.