Enhancement of Image Resolution using RDB

By using hierarchical features, Convolutional Neural Networks (CNNs) which is a Deep Learning method have shown notable improvements in image restoration (IR) tasks. However, poor performance arises from many existing deep CNN-based IR models' inability to adequately use the hierarchical information included in low-resolution images. To address the problem of image resolution, this study uses the Residual Dense Network (RDN) technology. RDN seeks to improve how well it uses hierarchical features from every convolutional layer. To be more precise, we have extracted localized information using the Residual Dense Block (RDB) and dense convolutional layers. By connecting all the layers of the present RDB twice the state of the preceding RDB, RDB creates a contiguous memory system. We incorporated Local Feature Fusion (LFF) into Residual Dense Blocks to improve the feature extraction from historical and current local data. This improved the stability of training for larger networks. After getting dense local properties, we used Global Feature Fusion (GFF) to extract global hierarchical features. Through the use of RDN in Image Super-Resolution (ISR), we have proven its efficacy. Our tests on real datasets show that RDN performs exceptionally well on ISR tasks in visual quality as well as quantitative measures.