Article Preview
TopIntroduction
The advent and growth of artificial intelligence, computer vision and biometric techniques stipulates the identification of a person through handwriting. Writer Identification is a process to perform a one-to-many search in a known sample database to identify the writer of handwritten query document as illustrated in figure 1. The interest of researchers towards writer identification has been growing rapidly from the past decades because of its ample applications in various areas such as forensic science, biometric recognition, historical document analysis etc. It is also considered as a branch of behavioral biometric which helps to identify a writer’s various characteristics. Due to this, it is a convenient way to uniquely identify the person who has written a piece of text. However, it is considered as a challenging task due to within-writer writing variation (Bin Abdl & Hashim, 2015) as the same writer can have different writing styles at the different time, with a different pen even because of varying emotional state.
Figure 1.
General writer identification procedure
Writer Identification is the branch of writer recognition that primarily includes writer identification and verification process. As mentioned previously, the writer identification is a multi-class problem in which one-to-many search is performed for query document with previously stored documents whose authentication is known (Sagar & Pandey, 2015; Yang, Jin & Liu, 2016; Dargon et al., 2019). However, the writer verification is a binary class problem. In this, it has been verified that whether the document is written by claimed writer or whether two documents are written by same writer or not (Chawki Djeddi, 2010; Hanusiak et al., 2011; Halder, Obaidullah & Roy, 2016; Adak, Chaudhuri & Blumenstein, 2019).
On the basis of capturing of data, identification system can be categorized as an online or offline (Jain & Doermann, 2011; Jain & Doermann, 2014) as shown in figure 2. The online data is associated with temporal information and is collected on a tablet, touchpad via stylus, mouse or electronic pen. In addition to this, online data also have other features such as pen pressure, pen up-down events. However, the offline data is static and stored in the scanned image form. Offline writer recognition is considered harder as compared to online because of the lack of sequential information (Xiong, 2016) and it contains only scanned images of handwritten data (Yang, Jin & Liu, 2016).
Figure 2.
Online vs. Offline data capturing