Writer identification: A comparative study across three world major languages

Area of Science:

• Computer Science
• Pattern Recognition
• Forensic Science

Background:

• Offline text-independent writer identification is crucial for forensic analysis.
• Existing literature reviews often focus on single languages or broader timeframes.
• A comparative analysis across major languages like English, Chinese, and Arabic is needed.

Purpose of the Study:

• To review and categorize state-of-the-art offline text-independent writer identification techniques.
• To compare the advancements in writer identification for English, Chinese, and Arabic languages between 2011 and 2016.
• To identify challenges and limitations in current methodologies, particularly concerning dataset size and language complexity.

Main Methods:

• Literature review of scientific papers published between 2011 and 2016.
• Categorization of writer identification techniques into texture-based, structure-based, and allograph-based methods.
• Comparative analysis of method performance across different languages and datasets.

Main Results:

• Significant progress was observed in writer identification for English and Arabic texts during the review period.
• Writer identification for Chinese texts showed slower progress compared to English and Arabic, attributed to its complex writing system.
• The accuracy of writer identification methods generally decreases with an increase in the size of the utilized dataset.

Conclusions:

• The state-of-the-art in writer identification has advanced, but disparities exist across languages.
• The complexity of the Chinese writing system presents a significant challenge for current identification techniques.
• Dataset limitations and scalability issues need to be addressed for more robust and accurate writer identification systems.