Segmentation Free FragNet and GRRNN for Writer-Identification in Cursive Handwriting

Abstract

Writer identification is a method in which a handwritten sample is assigned to the corresponding author of that sample. Since handwriting is a behavioral biometric application like the signature, writer identification will also confirm the authenticity of the document. Major applications of the writer identification process come in forensics, banking, and paleography which is a study of historical books and writings. Different writers have their own style of handwriting which has different characteristics, so they require different handling. Due to this difference, it is difficult to identify the author of the manuscript. Slant, curvature, orientations, etc may vary. The writer identification task predicts the writer by identifying various features involved in the handwrit- ten sample. The proposed method identifies the writer of a handwritten sample by combining both local and global features. This is accomplished by a combination of FragNet and Global Recurrent Neural Network. FragNet captures local features and to make them spatially correct, global features are captured from local features by Global Residual Recurrent Neural Network. A segmentation-free method is used so that writers of cursive handwriting can also be identified.