The paper introduces a novel residual dense network (RDN) for image super-resolution (SR) to address the issue of not fully utilizing hierarchical features from low-resolution (LR) images in deep CNN-based SR models. The RDN is designed to exploit these hierarchical features more effectively by proposing a residual dense block (RDB) that includes dense connected convolutional layers and local feature fusion (LFF). The RDB allows direct connections between preceding and current layers, enabling contiguous memory (CM) and enhancing the flow of information. Global feature fusion (GFF) is also introduced to adaptively learn global hierarchical features. Experiments on various benchmark datasets with different degradation models show that the RDN outperforms state-of-the-art methods, demonstrating its effectiveness and robustness. The main contributions of the paper are the unified framework of RDN, the RDB, and the GFF, which together achieve superior performance in image SR.The paper introduces a novel residual dense network (RDN) for image super-resolution (SR) to address the issue of not fully utilizing hierarchical features from low-resolution (LR) images in deep CNN-based SR models. The RDN is designed to exploit these hierarchical features more effectively by proposing a residual dense block (RDB) that includes dense connected convolutional layers and local feature fusion (LFF). The RDB allows direct connections between preceding and current layers, enabling contiguous memory (CM) and enhancing the flow of information. Global feature fusion (GFF) is also introduced to adaptively learn global hierarchical features. Experiments on various benchmark datasets with different degradation models show that the RDN outperforms state-of-the-art methods, demonstrating its effectiveness and robustness. The main contributions of the paper are the unified framework of RDN, the RDB, and the GFF, which together achieve superior performance in image SR.