A Simple Convolutional Neural Network for Small Sample Multi-lingual Offline Handwritten Signature Recognition

Abstract

AbstractBased on convolutional neural network, a simpler network for offline handwritten signature recognition is proposed in this paper. A total of 7,200 handwritten signature images in Chinese, Uyghur, and Kazakh languages were collected from 300 volunteers to establish an offline multi-lingual handwritten signature database. The signature images were randomly divided into training, verification, and testing sets. The average accuracy of five experiments was taken as the recognition result. Because it is difficult to collect a large number of signatures at one time, many experiments have been carried out on a multi-lingual signature database and CEDAR database with the training samples are 6, 12, and 18 respectively. Experimental results show that this method is an effective signature recognition method, and the effect is better than the existing network when there are few training samples.KeywordsMulti-lingualHandwritten signatureSignature recognitionConvolutional neural network