A Novel Method to Recognise Closely Connected CAPTCHA
Liang Zhang, Shu Guang Huang, Ronggui Hu, He Ping Tang, Zhao Xiang Shi · Advanced materials research · 2012
Nowadays, most of text CAPTCHAs use character connection as a primary means to avoid being recognized. To recognize this kind of CAPTCHA, a new image analysis model, Concept Component Analysis (CCA), is proposed. Based on the approaching idea in Newton’s iteration, this new model is solved by a multi-population genetic algorithm. Compared with traditional image analysis models, such as PCA, components obtained by CCA have obvious concept meanings. Images can be recognized by solely relying on these components, No classifier is needed. CCA has achieved good recognition results in our experiments. Suggestions for securing text recognition CAPTCHAs are also provided based on experiments results.