Image Super-Resolution Reconstruction Based on Holistic Attention Dual Regression

When addressing the challenges of an excessively large solution space and the difficulties in channel attention mechanisms to restore texture details during the generation of High-Resolution (HR) images from Low-Resolution (LR) images, an image super-resolution reconstruction algorithm based on the Holistic Residual Attention Dual Network (HRADN) is proposed. The use of a Dual Regression Network (DRN) with additional mappings is employed to estimate the downsampling kernel and impose constraints on the LR images, thereby reducing the solution space of functions. The incorporation of the Layer Attention Module (LAM) and Channel-Spatial Attention Module (CSAM) explores the correlations between different modules in the network and uncovers deeper information within the training data. Experimental evaluations conducted on five benchmark datasets have shown that the proposed algorithm outperforms several prevalent models in terms of Peak Signal-to-Noise Ratio (PSNR), indicating its superior performance in super-resolution reconstruction. It significantly improves the clarity of details and reconstructs sharper high-frequency textures in subjective visual effects.