Handwriting analysis of offline data for gender prediction

Significant progress has been made in the field of Writer identification and Signature Verification. Handwriting Analysis can be applied for classifying age, gender and nationality. Writer Recognition through handwriting analysis can be divided into two categories: online and off-line. On-line character recognition involves the identification of characters while they are written and deals with time ordered sequences of data, pen up, and down movement and pressure sensitive pads that record the pen’s pressure and velocity. On the other hand, off-line character recognition involves the recognition of already written character patterns in scanned digital image. Writer Recognition approaches can be categorized into two families: Text-Dependent writing and Text-Independent writing. In this research we were intended to improve the handwriting analysis performance of offline data using cursive text-independent Arabic and English writing for genders by proposing new features, algorithms and classification methods. We proposed four new features for the experiments i.e. LBP (Local Binary Pattern), SFTA (Segmented-based Fractal Texture Analysis), HOG (Histogram of oriented Gradient) and GLCM (Gray level Co-occurrence Matrix). We also performed some more experiments on the dataset including Text-independent, Text-dependent, Script-dependent and Script-independent. We performed comparison between our proposed approach and previous methods. All these experiments are performed on QUWI database. It contains 4068 handwritten texts by 1013 writers (male and female). We report a percentage of 81% improvement in the handwriting analysis for gender prediction of the offline data.