Dual-Pathway Deep CNN for Offline Writer Identification

Abstract Writer Identification is a challenging problem based on small amount of handwritten text. It is also a foremost research topic in forensic analysis of documents. For this, a convolution neural network (CNN) is proposed to address the writer recognition task. In this work, we propose a dual-path CNN to extract local features maps from given input handwritten document image patches and then classify writer by Softmax loss function. The experiments are done on IAM English language dataset and obtained accuracy of 92.7%.KeywordsWriter identificationDeep learningFeature extraction