Writer Identification using Deep Learning with FAST Keypoints and Harris corner detector

Abstract Writer identification from offline images of handwriting is an interesting pattern classification problem that has been investigated for many decades now. Despite significant research endeavors, the problem still remains challenging due to high intra-class variations and, at times, high similarity between writings of two individuals. This paper presents a writer identification system that relies on extraction of key points from handwriting and feeding small patches around these key points to a convolutional neural network for feature learning and classification. More specifically, we employ FAST key points and Harris corner detector to identify points of interest in the handwriting. A deep CNN is trained using small patches centered around these key points. Classification is carried out in an end-to-end manner as well as by encoding features extracted from one of the convolutional layers. Experimental study is carried out on three different datasets namely IAM, QUWI and IFN/ENIT and high identification rates are reported for each of these datasets.