Image Super-Resolution Reconstruction Based on Dual-Branch Channel Attention

Image super-resolution reconstruction is an important technique for converting low resolution images into high-resolution images. High resolution images can provide more information and are crucial for advanced visual tasks. However, traditional methods cannot restore image details, resulting in blurring and not meeting practical requirements. In response to this, this paper proposes a dual-branch channel attention residual network (DCARN) for image super-resolution reconstruction, which focuses on the problem of multi-frequency information fusion. The proposed method first improve multi-spectral channel attention by adopting group convolution and discrete cosine transform to adapt to images of different sizes. In order to fully utilize channel and multi-spectral channel attention, a dual-branch channel attention residual block (DCARB) is further designed. Experiments on multiple public datasets show that the proposed method achieves improved performance and performs well on image reconstruction in terms of both subjective and objective quality, with richer stripe texture details.