N-FTRN: Neighborhoods based fully convolutional network for Chinese text line recognition

摘要: The convolutional recurrent neural network is one of the most popular text recognition methods. Recurrent structures can extract long-term dependencies, but they are time consuming in computation compared with structures. We argue that Chinese line be performed based on neighbor rather than entire contextual information, and information extracted from neighborhoods should only a supplement to character regions. Therefore, we propose novel fully (N-FTRN). It first extracts character-level feature sequences lines, then uses residual blocks instead structure utilize information. A reshape layer applied enable recognize both vertical horizontal lines. Extensive experiments have been conducted validate efficiency effectiveness proposed network. Compared state-of-the-art methods, achieve comparable performances scene competition dataset (TRW) ICDAR 2015 much more compact models.