Dense Residual Network: Enhancing global dense feature flow for character recognition

Deep Convolutional Neural Networks (CNNs), such as Dense Network (DenseNet), have achieved great success for image representation learning by capturing deep hierarchical features. However, most existing network architectures of simply stacking the convolutional layers fail to enable them fully discover local and global feature information between layers. In this paper, we mainly investigate how enhance abilities DenseNet exploiting features from all Technically, propose an effective model termed Residual (DRN) task optical character recognition. To define DRN, a refined residual dense block (r-RDB) retain ability fusion original RDB, which can reduce computing efforts inner at same time. After features, utilize sum operation several r-RDBs construct new (GDB) imitating construction blocks adaptively learn in holistic way. Finally, use two design down-sampling size extract more informative deeper Extensive results show that our DRN deliver enhanced results, compared with other related models.