Features extraction for text detection and localization

With the widely use of digital image, text location/detection is considered as a key part of the image text information extraction system. We have proposed a robust system for text detection in images. The proposed scheme uses a set of characteristics which are combined to build a large ones, it´s based on texture features and connected components analysis. First, edge detection technique is used to detect all edges in the image, to preserve the most important structural features of that image by finding sharp contrasts, after that, we select all closed contours from detected edges. By the next we use texture information to discriminate text from non-text regions. Finally, we verify regions in order to reduce false alarms and to generate valid text regions. The evaluation is done on the challenging ICDAR 2003 robust reading and text locating database. The results show that our system can localize text of various font sizes and styles in complex background. Indeed the recall/precision curve of the proposed method proves that 0.95 Precision rate is obtained for recall rate value equals to 0.89 and the standard measure of quality f is equal to 0.92.