method: TextCohesion_12019-04-27

Authors: weijiawu

Description: TextCohesion is text detecter for scene textdetection especially for those with arbitrary shapes.TextChohesion splits a text instanceinto 5 key components: a Text Skeleton, and four Directional pixel Regions. Thesecomponents are easy to handle rather than directly control the entire text instance. Wealso introduce a confidence scoring mechanism to filter out the characters that are similarto texts. Our method can integrate text contexts intensively even grasp clues when it iswith very complex background. Experiments on challenging benchmarks demonstratethat TextCohesion clearly outperform state-of-the-art methods. It achieves an F-measureof84.6and86.3on Total-Text and SCUT-CTW1500 respectively