A novel database for automatic processing of Persian handwritten bank checks. (February 2018)

• Record Type: Journal Article
• Authors: Akbari, Younes; Jalili, Mohammad J.; Sadri, Javad; Nouri, Kazem; Siddiqi, Imran; Djeddi, Chawki
• Is Part Of: Pattern recognition. Volume 74(2018:Feb.)
• Page Start: 253
• Page End: 265
• Publication Date: 2018-02
• Subjects: Document analysis -- Persian handwritten checks -- Automatic check processing -- Handwritten database -- Optical Character Recognition (OCR) -- Signature verification; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2017.09.011
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: A novel database for automatic processing of Persian handwritten bank checks. Database includes images of filled forms and checks with signatures. Ground truth allows evaluation of segmentation and recognition of check fields. Forged signatures allow evaluation of signature verification with skilled forgeries. Abstract: This paper introduces a database of Persian handwritten bank checks. The database includes legal amounts, courtesy amounts, dates, receiver names, signatures and account numbers. In addition to checks, the database also comprises handwritten forms with words used in legal amounts, digits employed in the courtesy amounts and signatures of contributors. The database can be employed for evaluation of segmentation and recognition of different fields in checks as well as for verification of signatures. Data is collected and organized in two series. The first series, comprising 500 hand filled Persian checks and 500 forms contributed by 500 different individuals, supports evaluation of segmentation and recognition tasks. The second series contributed by 100 writers supports evaluation of signature verification systems and comprises 200 genuine checks, 100 forms and 200 forged checks. To provide an insight to other researchers, experiments have been carried out on recognition of dates and courtesy amounts with recognition rates of 76% and 73%, respectively. Likewise, experiments on signature verification with skilled forgeries realized an average error rate of 12.25%. The database, named Persian Handwriting Bank Checks (PHBC) Database, is freely available to the research community upon request to the authors. It is expected that the developed database will not only allow researchers working on automatic processing of Persian checks to evaluate their proposed techniques but will also serve to meaningfully compare different recognition and verification techniques.