<p>This paper introduces the Composite Downsampling Model Based on Wavelet Transform and Bicubic Interpolation (CDWB), an innovative technique devised to effectively reduce the loss of high-frequency information in images that is caused by the actively down sampling operation. This model is pivotal in enhancing the capabilities of neural network algorithms to reconstruct high-frequency image details, thereby significantly improving the perceptual quality of the images. Utilizing the image data generated by our CDWB model in the training of super-resolution reconstruction networks yields high-resolution images that effectively replicate the pixel distribution of actual images. These images demonstrate a superior perceptual quality compared to those reconstructed from datasets trained with standard single interpolation-based downsampling techniques.</p>