Text Localization in Scene Images with SURF and Support Vector Machine for Multilingual Text

摘要

Detection of text is challenging due to complexities in the scene images with text and newer applications that use them. In our work we use general image repository that contains images suitable for various applications. We use Maximally Stable Extremal Regions (MSER)with Speed Up Robust Features (SURF)to find features. For training and testing Support Vector Machine (SVM)is used. Connected Component Analysis and filtering rules are used to get final localized text from the scene images. To evaluate our work, we have used existing data sets for English text and our own dataset for multilingual text.