Multi language text detection using fast stroke width transform

In few decades, text detection methods have been proposed with significant results. Yet, most of them are only implemented for certain language. Thus, this work proposes a text detection method for multi language text. To do such a task, a Fast Stroke Width Transform (FSWT) is introduced. FSWT is a robust local image operator for text detection with complex background. It is improved version of Stroke Witdh Transform (SWT) in order to obtain faster computation. The FSWT calculates the stroke width on each pixel. The pixels in the FSWT image connected into components according to stroke width value. These components are then classified as text or nontext. In standard SWT, the components should be generated two times according two types of text, bright text on dark background (BoD) and dark text on bright background (DoB). However, the FSWT could automatically determine the type of text by utilizing edge component labeling and tree of edge component. To compare with standard SWT, most of the images are archived in database with BoD and DoB text. In the experiment, our method achieves almost three times faster than the standard SWT in computation time. In consequence, our method is expected to be implemented with faster process in real-time case.