Developing an Empirical Algorithm for Protecting Text-Based CAPTCHAs against Segmentation Attacks

In this paper, we aim to provide an effective and efficient method to generate text-based Captchas which are resilient against segmentation attack. Different to the popular industry practice of using very simple color schemes, we advocate to use multiple colors in our Captchas. We adopt the idea of brush and canvas when coloring our Captchas. Furthermore, we choose to use simple accumulating functions to achieve diffusion on painted colors and DES encryption to achieve a good level of confusion on the brush pattern. To facilitate ordinary users and developers, we propose an empirical algorithm with support of Taguchi method to guarantee the quality of the chosen color schemes. Our proposed methodology has at least three advantages -- 1) the settings of color schemes can be fully customized by the user or developer, 2) the quality of selected colors have desirable statistical features that are ensured by Taguchi method, 3) the algorithm can be fully automated into computer programs. Moreover, our included examples and experiments prove the practicality and validity of our algorithm.