Offline Writer Identification and Verification Evaluation Protocols for Spanish Database

Writer identification based on text images is a well studied topic in Biometrics. Several methods have been proposed for this task. Despite the results achieved, current methods are limited in their ability to handle diverse languages, writing styles, and document types. In this work, we proposed relevant verification and open/closed-set evaluation protocols to assess the performance of writer identification methods on CENATAV-HTR dataset, containing Spanish handwritten documents. Under these evaluation protocols, we evaluate and analyze the effectiveness of a state-of-the-art Recurrent Neural Network originally proposed for English writer identification. The obtained results demonstrated that the models trained on English are not suitable to recognize writers in Spanish and thus they need to be adjusted or finetuned for this particular language. All text images and evaluation protocols are made available for future research in this topic.