Text localization in scene images is a challenging problem, and a lot of methods have been proposed for ICDAR dataset. Character Energy is a naive feature developed by observing images of characters in previous researches. However, the accuracy obtained when Character Energy is used depends on the accuracy of closed boundary detection. In this paper, we propose to combine using Character Energy and detecting MSERs. That is, we detect virtually unchanged regions with MSERs and compute Character Energy with the boundaries of detected regions. By using our proposal, we would properly extract text locations in scene images of ICDAR dataset.

Using the Hadoop architecture to propose a new theoretical model for brain system.

Social networking services (e.g., Twitter, Facebook) are now a major source of World Wide Web (called “Web”) dynamics. We investigate cascading mechanism in Twitter using transfer entropy.