Resnet-based slide puzzle captcha automatic response system

Slide puzzle captcha is a kind of dynamic cognitive game, which requires users to pass a series of cognitive tasks to verify themselves. Compared to boring text captcha, the user experience has been greatly improved, so slide puzzle captcha has gradually replaced the text-based captcha on many large platforms.In this paper, we divided slide puzzle captcha cracking into three steps: identifying the gap position, generating the sliding track, and implementing the browser automation. For the location identification of the gap, we used residual network based on object detection and yolov3-based object detection, establish Resnet-18 model and Yolov3 model, and in order to train the two models, we collect 1000 images from Bilibili, Netease Shield, Tik Tok, Jingdong, etcand estimated accuracy of gap identification; As for the generation of sliding trajectory, we analyze the sliding trajectory of human and imitated the human slider trajectory by the piecewise curve fitting of least-squares method; For the automatic implementation of browser, we calculate the offset position, use the TencentAPI, directly feed the recognition result to the page. We choose the resnet-18 and Yolov3 model to identify the location of the gap. We utilize the least-squares method to fit the sliding trajectory segmentally, increasing the degree of simulation and avoiding machine detection.