Semi-supervised feature learning for improving writer identification

Data augmentation is typically used by supervised feature learning approaches for offline writer identification, but such approaches require a mass of additional training data and potentially lead to overfitting errors. In this study, a semi-supervised feature learning pipeline is proposed to improve the performance of writer identification by training with extra unlabeled data and the original labeled data simultaneously. Specifically, we propose a weighted label smoothing regularization (WLSR) method for data augmentation, which assigns a weighted uniform label distribution to the extra unlabeled data. The WLSR method regularizes the convolutional neural network (CNN) baseline to allow more discriminative features to be learned to represent the properties of different writing styles. The experimental results on well-known benchmark datasets (ICDAR2013 and CVL) showed that our proposed semi-supervised feature learning approach significantly improves the baseline measurement and perform competitively with existing writer identification approaches. Our findings provide new insights into offline writer identification.