Handwriting Authentication Based on Pattern Recognition Using Deep Learning

Handwriting is a delicate human skill that can serve as a means of authenticating an author by reflecting unique behavioral traits. Document authentication as a practical application in the real world is challenging when faced with extracting handwriting style based on limited handwriting data and intra-class variations for each author. This article presents a novel deep learning-based framework for authenticating the author of a manuscript. The proposed writer identification framework is independent of the handwritten text and is applicable based on a small amount of data, such as one handwritten word. In order to obtain discriminative handwriting features, a deep hybrid structure is designed based on feature fusion. Furthermore, the application of a multiple loss function during the training phase represents an attempt to achieve a more accurate distinction between the handwriting styles of different authors. The efficacy of the proposed model is corroborated by experimental findings derived from three prominent datasets in English, Arabic and Dutch: IAM, KHATT and Firemaker. These results substantiate the model's performance across diverse linguistic contexts.