CapsNet regularization and its conjugation with ResNet for signature identification
作者：
Highlights：
• We proposed a regularized version of CapsNet aiming for more generalization and overfitting avoidance.
• Adding regularized term leads to removing the decoder in the baseline CapsNet that causes a noticeable reduction of learning parameters and fast convergence.
• We also proposed a conjugation of regularized CapsNet with ResNet dealing with the small number of training samples or the input images with a large size.
• We evaluated our approach over three well-known publicly available datasets.
摘要
•We proposed a regularized version of CapsNet aiming for more generalization and overfitting avoidance.•Adding regularized term leads to removing the decoder in the baseline CapsNet that causes a noticeable reduction of learning parameters and fast convergence.•We also proposed a conjugation of regularized CapsNet with ResNet dealing with the small number of training samples or the input images with a large size.•We evaluated our approach over three well-known publicly available datasets.
论文关键词：Regularized capsule neural network,Residual neural network,CapsNet regularization,CapsNet and ResNet conjugation,Signature recognition
论文评审过程：Received 26 August 2020, Revised 9 December 2020, Accepted 24 January 2021, Available online 29 January 2021, Version of Record 10 July 2021.
论文官网地址：https://doi.org/10.1016/j.patcog.2021.107851
