Abstract: In order to effectively recognize the information of text by using Android intelligent terminals, we proposed a text feature extraction method based on spectral residual (SR) and stroke width transform (SWT). Firstly, the salient map was obtained based on salient detection of the spectral residual. Secondly, the maximum entropy discriminant method was used to segment and establish salient text candidate regions in salient map. Finally, the SWT algorithm was used to extract characteristics of equivalent width of closed edge in the candidate region. The C/S architecture experimental platform was constructed based on Android intelligent terminal and server, and experimental results on both the MSRA-TD500 text dataset and laboratory dataset show that the algorithm can effectively obtain salient text feature in the image, and the performance of the algorithm is improved when the computing efficiency is guaranteed.

Key words: spectral residual, text feature detection, Android terminal, stroke width transform