Automated text detection and text-line construction in natural images

This work develops an automated system to detect texts in natural images captured by the cameras embedded on mobile devices. Unlike former researches which focus on detecting with straight texts, this work proposes a text-line construction algorithm which is able to extract curved text-lines in any orientations. An image operator called the Stroke Width Transform is adopted to exploit connected components which have stroke-like properties. Text components are classified into two types: active and passive. The links of active components are considered the initial orientation of text-lines. Complete text-lines are constructed by linking active and passive components. The system is implemented on the Android platform and the experimental results demonstrate the feasibility and validity of the proposed system.