Abstract:
"Handwriting is a sort of behavioral human biometrics that is unique to each individual and
changes over time. It can be used as a secondary identifying mechanism for systems that require
verification or in legal proceedings. Different factors such as the writing tool used, the type of
paper used, the ink flow, and the pressure used when writing all contribute to a person's
handwriting, making it difficult to manually identify the writer of an unidentified document.
Because handwriting can be copied in circumstances of litigation requiring writer
identification, an accurate assessment is critical.
This research project attempts to automate writer identification by following a novel deep
learning approach in order to further improve accuracy in predictions compared to already
existing systems. Even though there are current existing implementations that follow a deep
learning approach, they do not implement solutions that utilizations feature extraction methods.
Additionally, the author attempts to create a standardized dataset of the Sinhala language since
none exists and also it is recognized as a low resource language which further validates the
need of creating a dataset.
To conclude this research will greatly contribute both the technical and domain aspect of writer
identification as the development of a novel deep learning approach and the creation of a
standard dataset of a low resource language will be beneficial to the field. The proposed
solution can also be used as a method for detecting perjury in litigation."
