Authors: Jonathan Fabrizio, Myriam RobertSeidowsky, Edwin Carlinet, Thierry Geraud

Description: This method has two main steps: a hypothesis generation step to get potential text boxes and a hypothesis validation step to filter false detections. The hypothesis generation step process relies on the TMMS segmentation method, while the validation step is based on a texture-based SVM classification approach.
This variant includes a split algorithm in order to match the GT granularity (word level).