On the detection of textual information in metro stations
Muhammad Shehzad Hanif, Lionel Prévost · 2009
We present a robust and efficient method for extracting textual information in the metro and train stations. The textual information in the train stations is destined to guide passengers about the directions, name of the station, etc. Our proposed method for text detection is based on the adaptive boosting method (AdaBoost algorithm) to construct a precise classifier by selecting and combining several features and weak classifiers of different families. We have studied, in detail, the behavior of the particular boosting algorithm when several features and weak classifiers are employed in the feature selection process. The evaluation is done on a challenging metro station image database which contains text of various sizes and fonts. The experiments show that the text detector is robust at detecting text in complex backgrounds.