A Crowdsourcing-based Image Caotcha Assistance System For Visual Impairment

As an efficient test for human-computer distinction, image CAPTCHAs have prevented automatic program such as web bots and scripts from abusing online resources and services, so it is widely used in ticket-booking system, bulletin board system, blogs and social networks. However, it has become the greatest obstacle for the blind to surfing the web. Since screen-reader software is only designed for standard text information and present image recognition technologies fail to recognize complicated image CAPTCHAs accurately, a crowdsourcing-based CAPTCHA recognition assistance system is one of the best solutions to help the blind to skip CAPTCHAs test.The main goal in this paper is to design and implement an image-based CAPTCHAs recognition assistance system for the blind. The server side uses server cluster instead of single server, which increases the concurrency effectively, and at the same time, decrease the probability of service unavailability due to single server failure. Considering the difference in the performance and load among the server nodes, the system employs dynamic load balancing algorithm to improve the server utilization rate and reduce loss in performance caused overload.Since CAPTCHAs play an important role in the web security, it is necessary to take actions to prevent the users with malicious intent from abusing the service to skip the CAPTCHAs for network attack. So a CDPF certification and real-name authentication are required for a registration, In addition, the system constructed a user behavioral model for anomaly behavioral detection and will stop providing service for anomaly behavioral users. At the same time, the system employs single sign-on unified identify authentication, which makes the user unnecessary to login for times for different type of services.Moreover, the system optimize the request queue scheduling algorithm according to the timeliness of the CAPTCHAs and the result of previous request, which obviously prevent some users for waiting for a long time and reduce the probability of service timeout and improved the average users’experience. Besides, the system also using probability theory to analyze and calculate to determine the best number of staff required to reduce waste of human resources.Finally, through testing and analysis on the server side, the system designed and implemented in this paper has good expansibility; it is convenient to add additional functions such as blind online game and blind chatting to present system to construct a relatively comprehensive-functional information accessibility platform for the blind.