Japanese Writer Identification Method Introducing Attention Mechanism into ResNet to Capture Local Writing Habits

Writer identification is a task to estimate the person who wrote a character in the input character image by estimating whose writing habits are similar to those of writers in the database. In recent years, the number of writer identifiers based on neural networks has been increasing. However, they have not been able to capture local features. Therefore, we proposed to introduce an attention mechanism into ResNet. The experimental results of changing the position of the attention mechanism made it possible to capture local features.