Comparative study on language independent forensic writer identification

Offline writer identification is widely applied in various research areas. Forensic document analysis is a typical motivation in which researchers strive for the best possible performance using limited amount of information. This paper is concerned with offline, text-dependent writer identification. Three techniques are analyzed and evaluated on different levels of analysis using different scripts. The first technique is based on geometric moments. The second technique relies on signature matching. And the third technique is derived from fractal analysis. Several tests are performed on both English and Arabic texts on the phrase level, the word level and the character level. The best identification accuracy reached a top-1 result of 94% and 99.6% on a database of 50 writers using selected word and character data sets respectively. The performance of the system was also tested on selected subsets from the IAM English handwriting benchmark database.