Global vs. local features for gender identification using Arabic and English handwriting

The focus of this paper is off-line analysis of handwriting for the purpose of identifying a writer's gender. An example application of such identification is forensic analysis, in which gender could be combined with other biometric attributes for identification or exclusion. Studies have shown that handwriting by males and females tend to exhibit distinctive characteristics, even across different languages and cultures. We have explored these differences by developing classifiers based on support vector machines (SVM). These were trained using a database in which 282 individuals provided handwriting samples in both Arabic and English. The paper compares results obtained using local features and global features. For classifiers using global features trained using handwriting samples for both languages, an accuracy of 81% was observed. When local features were also used, however, an accuracy of 94.7% was observed. These results highlight the importance of local feature analysis for gender identification from handwriting.