Handwriting based writer recognition using implicit shape codebook

Writer characterization from images of handwriting has remained an important research problem in the handwriting recognition community that finds applications in forensics, paleography and neuropsychology. This paper presents a study to evaluate the effectiveness of an implicit shape codebook technique to recognize writer from digitized images of handwriting. The technique relies on identifying the key points in handwriting and clustering the patches around these key points to generate an implicit shape codebook. A writer is then characterized by the probability distribution of producing the codebook patterns. Experiments are carried out in text-dependent as well text-independent mode using the standard BFL and CVL databases of handwriting images. Promising identification and verification performance is reported in a number of interesting experimental scenarios.