A Novel Joint Character Categorization and Localization Approach for Character-Level Scene Text Recognition

摘要: Scene text recognition has become an active research area in pattern recent years. Currently, the mainstream approach is image-based sequence model. However, such a model usually cannot yield accurate character-level category and location information. To address this deficiency, paper, we propose novel scene framework for simultaneously categorizing localizing characters. Moreover, present effective joint learning strategy to help learn from both annotation word-level annotation. Extensive experiments on five benchmark data sets, including IIIT-5K, SVT, ICDAR03, ICDAR13, ICDAR15, show promising results. Especially, confirm that our proposal more robust length variation non-language text.