Image dehazing is a specific image restoration task that aims to recover the quality and clarity of a haze image caused by ambient phenomena. In this paper, we present a novel transformer-based approach for image restoration and texture enhancement. Our model adopts novel Reference Super-Resolution (RefSR) deep models capable of transferring high-resolution texture information from a reference image to a blurred image, and we named it TTDN. Our approach employs a VGG19-based network for deep feature extraction and then unfolds these features into patches. Then, we incorporate textures at multiple scales, enhanced by gradient density information, into a cohesive reconstruction through a simple residual network. Experimental results demonstrate that our approach achieves competitive state-of-the-art performance for both peak signal-to-noise ratio (PSNR) and structure similarity index (SSIM) metrics. Moreover, our approach offers a significant contribution to the field by adapting the RefSR paradigm and improving the computational feasibility of patch-based texture transfer. Experimental results in the test RESIDE-indoor data set show competitive metric results that other SOTA methods while improving efficiency in large-scale images by using larger patches with larger strides. Our results suggest that the proposed transfer strategy can be applied to other image restoration tasks, offering a promising avenue for future research.