A Text Detection Technique Applied in the Framework of a Mobile Camera-Based Application

Recent advances in mobile devices allow us to address many new challenging problems. One of them is automatic text recognition for embedded platform. This paper describes an innovative text detection system in the context of an embedded camera-based application. We propose a method to identify text regions inside an image, to correct orientation problems and to analyze document layout. Text areas are isolated with a texture segmentation approach. Due to mobile conditions, text orientation and perspective must be corrected. First a fuzzy estimation of text orientation is computed quickly. If text is too much distorted, the text perspective is corrected by using a line segmentation method in two steps. Finally the layout of the document is computed in order to deliver the reading order of the document. This language-free system has been developed with special attention to computational performances. The experimental results have proven that the method is effective and realistic.