文摘
As the scene text detection and localization is one of the most important steps in text information extraction system, it had been widely utilized in many computer vision tasks. In this paper, we introduce a new method based on the maximally stable extremal regions (MSERs). First, a coarse-to-fine classier estimates the text probability of the ERs. Then, a pruning algorithm is introduced to filter non-text MSERs. Secondly, a hybrid method is performed to cluster connected components (CCs) as candidate text strings. Finally, a fine design classifier decides the text strings. The experimental results show our method gets a state-of-the-art performance on the ICDAR2005 dataset.