基于细节增强框架的图像超分辨率算法

Image super-resolution algorithm based on detail-enhancing framework

基于参考图像的图像超分辨率(Ref-SR)技术将高分辨率(HR)参考图像导入单输入图像超分辨率(SISR)方法中,利用参考图像中转移的纹理缓解了此方法中长期存在的不适定问题。尽管定量和定性结果的显著提升已经验证了Ref-SR方法的优越性,但在纹理转移之前存在的不对齐问题表明该方法中还有进一步提高性能的空间。现有方法忽视了在对齐特征时细节的重要性,因此没有充分利用低分辨率(LR)图像中包含的信息。文中提出一种基于细节增强框架(DEF)的参考图像超分辨率方法,该方法引入扩散模型来生成和增强LR图像中的潜在细节。如果参考图像中存在对应部分,设计方法可以使得图像特征的对齐更为准确;在参考图像缺少对应部分的情况下,LR图像中的细节仍然得到了增强,同时避免了参考图像转移错误纹理的影响。大量实验表明,提出的方法相比于以往方法在数值结果相仿的前提下取得了优越的视觉效果。

In the technology of the reference-based image super-resolution(Ref-SR),the high-resolution(HR)reference images are imported into the single-input image super-resolution(SISR)approach,and the long-standing ill-posed issue of the very method has been alleviated with the assistance of texture transferred from reference images.Although the significant improvement in quantitative and qualitative results has verified the superiority of Ref-SR methods,the presence of misalignment before texture transfer indicates room for further performance improvement.Existing methods tend to neglect the significance of details in the context of comparison,so they fail to leverage the information contained within low-resolution(LR)images.In view of the above,a reference-based image super-resolution method on the basis of detail-enhancing framework(DEF),which introduces the diffusion model to generate and enhance the underlying detail in LR images,is proposed.If corresponding parts are present in the reference images,the proposed method can facilitate rigorous alignment.In cases where the reference images lack corresponding parts,the details in the LR images are still enhanced,and the influence of the wrong texture transferred from the reference images is avoided.Extensive experiments demonstrate that the proposed method achieves superior visual results while maintaining comparable numerical results in comparison with the existing methods.