Wide Activation Fourier Channel Attention Network for Super-Resolution

Attention mechanisms, especially channel attention, have been widely used in a wide range of tasks in computer vision. More recently, researchers have begun to apply channel attention mechanisms to tasks involving single image super-resolution (SISR). However, these mechanisms, borrowed from other computer vision tasks, may not be well-suited for SISR, which primarily focuses on re-covering high-frequency information. Consequently, existing approaches may not adequately reconstruct high-frequency details. To address this limitation, we propose a novel channel attention block, i.e., the Fourier channel attention block (FCA). This block leverages the Fourier transform to extract high-frequency information and subsequently compresses the spatial information, thereby emphasizing the high-frequency components within the image. To further enhance the performance, we propose a wide activation Fourier channel attention super-resolution network (WFCASR) to enhance the residual block by incorporating the wide activation mechanism and FCA. Results in the development of. By integrating the FCA block and the wide activation mechanism into our network, the high-frequency information can be effectively reconstructed and thus the accuracy and effectiveness of SISR can be effectively improved. Experimental results demonstrated that Our FCA channel attention mechanism has better performance.