Single Image Super-Resolution Reconstruction Technique with Lightweight Dual-Branch Network
Ruixiong Gu, Zhijun Shen, Junjie Chen · 2023
With the development of convolutional neural networks, the performance of Single Image Super-Resolution (SISR) has been significantly improved. However, as these methods are inevitably accompanied by complex operation and performance consumption issues, it is difficult to apply in practice. This paper proposes a lightweight Dual-Branch network (DBNet) for SISR, which can achieve a good balance between model complexity and performance. DBNet adopts a new intermediate module, Feature Extraction Dual Attention Block (FERDB) for feature extraction. DBNet combines Backward Fusion Module (BFM) and recursive Transformer. A new Attention mechanism, Multipath Attention (MA) is used to obtain key feature information. In addition, DBNet uses a Dual-Branch structure to capture more characteristic information. The experimental results show that DBNet is superior to other methods in SISR performance, with relatively lower complex operation and performance consumption.