Effectiveness of feature weight using BPSO in text-dependent writer identification

Abstract

Writer identification is an authorship authentication process based differences and similarities in handwriting. The main issue in writer identification is how to get the features that invariant to the writer. This study proposes Binary Particle Swarm Optimization (BPSO) based off-line text-dependent to investigate the effectiveness of feature weight in writer identification. BPSO has ability to perform such role since BPSO works on particle level and swarm level. The weight obtained by BPSO it's an average of feature selected times over 10 runs per writer. Then each feature multiplied by its corresponding weight so the features represented by their importance not their values. Off-line text-dependent words from IAM database are used. Moment and statistical features are extracted to represent the handwritten words. Experimental results show an improvement in writer identification performance based feature weight.